---
apiVersion: v1
items:
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.38/23"],"mac_address":"0a:58:0a:84:00:26","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.38/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.38"
            ],
            "mac": "0a:58:0a:84:00:26",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:39:28Z"
    generateName: auth-disabled-test-kserve-5f858fd8bc-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload
      app.kubernetes.io/name: auth-disabled-test
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: both
      pod-template-hash: 5f858fd8bc
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"15faa112-17a8-4fd2-88a3-232ea464670f"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.38"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:38Z"
    name: auth-disabled-test-kserve-5f858fd8bc-gcn7h
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: auth-disabled-test-kserve-5f858fd8bc
      uid: 15faa112-17a8-4fd2-88a3-232ea464670f
    resourceVersion: "30841"
    uid: 2763ae5e-3cc2-45cc-9c05-53458eed0b56
  spec:
    containers:
    - command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" "publishers/kserve-ci-e2e-test/models/facebook/opt-125m" \
          --port 8000 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 10
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 2
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 1
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-vqh4l
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-vqh4l
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: default
    serviceAccountName: default
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - emptyDir: {}
      name: tmp-dir
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: auth-disabled-test-kserve-self-signed-certs
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-vqh4l
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:29Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:38Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:38Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:38Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:28Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://9b030813fc50cf55b70def05f3740aa1b25a1ce3146a00550c040782c9a31007
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:39:38Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-vqh4l
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://73237a1bc987b3a5a44c6bd06221295c99aadf17ac0e538bd0b4430ac09f8b7d
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://73237a1bc987b3a5a44c6bd06221295c99aadf17ac0e538bd0b4430ac09f8b7d
          exitCode: 0
          finishedAt: "2026-07-02T21:39:37Z"
          reason: Completed
          startedAt: "2026-07-02T21:39:28Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-vqh4l
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.38
    podIPs:
    - ip: 10.132.0.38
    qosClass: Burstable
    startTime: "2026-07-02T21:39:28Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      app.kubernetes.io/version: 0.9.0
      certificates.kserve.io/expiration-v2: "true"
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.39/23"],"mac_address":"0a:58:0a:84:00:27","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.39/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.39"
            ],
            "mac": "0a:58:0a:84:00:27",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:39:28Z"
    generateName: auth-disabled-test-kserve-router-scheduler-7dcc8bd5f8-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-router-scheduler
      app.kubernetes.io/name: auth-disabled-test
      app.kubernetes.io/part-of: llminferenceservice
      pod-template-hash: 7dcc8bd5f8
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:app.kubernetes.io/version: {}
            f:certificates.kserve.io/expiration-v2: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"3accda8d-f91f-4507-b5b9-a0b664cba0b3"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":5557,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9002,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9003,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/tmp/tokenizer"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"STORAGE_ALLOW_PATTERNS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tokenizer-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-tmp"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-uds"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:39:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.39"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:40:02Z"
    name: auth-disabled-test-kserve-router-scheduler-7dcc8bd5f8-2tczn
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: auth-disabled-test-kserve-router-scheduler-7dcc8bd5f8
      uid: 3accda8d-f91f-4507-b5b9-a0b664cba0b3
    resourceVersion: "29278"
    uid: 8e9b9dea-521f-4c40-9793-d06edd129d21
  spec:
    containers:
    - args:
      - --config-text
      - |
        apiVersion: inference.networking.x-k8s.io/v1alpha1
        kind: EndpointPickerConfig
        plugins:
        - type: single-profile-handler
        - type: queue-scorer
        - type: prefix-cache-scorer
        - type: max-score-picker
        schedulingProfiles:
        - name: default
          plugins:
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
      command:
      - /app/epp
      - --pool-name
      - auth-disabled-test-inference-pool
      - --pool-namespace
      - kserve-ci-e2e-test
      - --zap-encoder
      - json
      - --grpc-port
      - "9002"
      - --grpc-health-port
      - "9003"
      - --enable-cert-reload=true
      - --secure-serving=true
      - --model-server-metrics-scheme=https
      - --cert-path=/var/run/kserve/tls
      env:
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: liveness
        initialDelaySeconds: 5
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 9002
        name: grpc
        protocol: TCP
      - containerPort: 9003
        name: grpc-health
        protocol: TCP
      - containerPort: 9090
        name: metrics
        protocol: TCP
      - containerPort: 5557
        name: zmq
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: readiness
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mwp9x
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: auth-disabled-test-epp-sa-dockercfg-qgsj9
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: STORAGE_ALLOW_PATTERNS
        value: '["tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
          "vocab.json", "merges.txt", "config.json", "generation_config.json"]'
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mwp9x
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: auth-disabled-test-epp-sa
    serviceAccountName: auth-disabled-test-epp-sa
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: auth-disabled-test-kserve-self-signed-certs
    - emptyDir: {}
      name: tokenizer-uds
    - emptyDir: {}
      name: tokenizer-tmp
    - emptyDir: {}
      name: tokenizer-cache
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-mwp9x
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:29Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:30Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:40:02Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:40:02Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:39:28Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 256m
        memory: 500Mi
      containerID: cri-o://5c832edafe19f8bafae27cad6c915d602cb31f37130d8998e962abc0d58aca08
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-endpoint-picker@sha256:06b6c75d77afd0e07053402752a9736c2dfbc12a306d0d37d963aac4c1d4e6a6
      lastState: {}
      name: main
      ready: true
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:39:30Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mwp9x
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://807f84fd05a1a63925e7050ffa27faeeca61be9aade0d03063ad935851e065ac
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://807f84fd05a1a63925e7050ffa27faeeca61be9aade0d03063ad935851e065ac
          exitCode: 0
          finishedAt: "2026-07-02T21:39:29Z"
          reason: Completed
          startedAt: "2026-07-02T21:39:28Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mwp9x
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.39
    podIPs:
    - ip: 10.132.0.39
    qosClass: Burstable
    startTime: "2026-07-02T21:39:28Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.48/23"],"mac_address":"0a:58:0a:84:00:30","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.48/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.48"
            ],
            "mac": "0a:58:0a:84:00:30",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:10:51Z"
    generateName: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-76cbf745cc-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-50bc673d
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: decode
      pod-template-hash: 76cbf745cc
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"c2a652c2-5109-4953-b97c-f1b61d1d9603"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8001,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"llm-d-routing-sidecar"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"INFERENCE_POOL_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"INFERENCE_POOL_NAMESPACE"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources: {}
              f:restartPolicy: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.48"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:13:51Z"
    name: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-76cbf745cc72pxw
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-76cbf745cc
      uid: c2a652c2-5109-4953-b97c-f1b61d1d9603
    resourceVersion: "62688"
    uid: 7276a9b3-d49d-42a0-b5b8-9944bc6f0974
  spec:
    containers:
    - command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" "publishers/kserve-ci-e2e-test/models/facebook/opt-125m" \
          --port 8001 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 8
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 180
        periodSeconds: 30
        successThreshold: 1
        timeoutSeconds: 30
      name: main
      ports:
      - containerPort: 8001
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8001
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-dockercfg-4wf5m
    initContainers:
    - command:
      - /app/pd-sidecar
      - --port=8000
      - --vllm-port=8001
      - --kv-connector=nixlv2
      - --enable-ssrf-protection=true
      - --pool-group=inference.networking.x-k8s.io
      - --inference-pool=kserve-ci-e2e-test/llmisvc-model-fb-opt-125m-route-50bc673d-inference-pool
      - --secure-proxy=true
      - --cert-path=/var/run/kserve/tls
      - --decoder-use-tls=true
      - --prefiller-use-tls=true
      env:
      - name: INFERENCE_POOL_NAMESPACE
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.namespace
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      - name: INFERENCE_POOL_NAME
        value: llmisvc-model-fb-opt-125m-route-50bc673d-inference-pool
      image: ghcr.io/llm-d/llm-d-router-disagg-sidecar:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      livenessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 10
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 10
      name: llm-d-routing-sidecar
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 10
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 10
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources: {}
      restartPolicy: Always
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-fb-opt-125m-route-50bc673d-kserve
    serviceAccountName: llmisvc-model-fb-opt-125m-route-50bc673d-kserve
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir: {}
      name: tmp-dir
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv44d181485fad85e662eb092f3749502f-kserve-self-signed-certs
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-5ccvl
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:52Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:57Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:13:51Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:13:51Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:51Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://1176b8ee20f97eccf7e93279b2075475f1c34b0bb92a6bc8d75263ab43c848f1
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:10:57Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - containerID: cri-o://d019f5d6f8f9b76d72b6d2b643edbd59a87926c5b6537aacd773a370172e092b
      image: ghcr.io/llm-d/llm-d-router-disagg-sidecar:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-disagg-sidecar@sha256:1ec2c035216ea0a4f8a6f7ccda55398fc3d9f71f487c135f49ba0dea08163f12
      lastState: {}
      name: llm-d-routing-sidecar
      ready: true
      resources: {}
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:10:51Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
        recursiveReadOnly: Disabled
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://6ebdbc351158f9ae8938d23fe845486a5353fac37566d855a2ab40f1dde505dd
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://6ebdbc351158f9ae8938d23fe845486a5353fac37566d855a2ab40f1dde505dd
          exitCode: 0
          finishedAt: "2026-07-02T22:10:56Z"
          reason: Completed
          startedAt: "2026-07-02T22:10:52Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-5ccvl
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.48
    podIPs:
    - ip: 10.132.0.48
    qosClass: Burstable
    startTime: "2026-07-02T22:10:51Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.49/23"],"mac_address":"0a:58:0a:84:00:31","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.49/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.49"
            ],
            "mac": "0a:58:0a:84:00:31",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:10:51Z"
    generateName: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-prefill-649c8bb55d-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload-prefill
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-50bc673d
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: prefill
      pod-template-hash: 649c8bb55d
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"80dea024-6aad-4f29-97d7-98a53ca79173"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.49"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:13:51Z"
    name: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-prefill-64sg2zn
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-route-50bc673d-kserve-prefill-649c8bb55d
      uid: 80dea024-6aad-4f29-97d7-98a53ca79173
    resourceVersion: "62701"
    uid: 67b939e3-31fa-4353-90ba-bd4834ed659c
  spec:
    containers:
    - command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" \
          --port 8000 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 8
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 180
        periodSeconds: 30
        successThreshold: 1
        timeoutSeconds: 30
      name: main
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: File
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-csgnh
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-csgnh
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: default
    serviceAccountName: default
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir: {}
      name: tmp-dir
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv44d181485fad85e662eb092f3749502f-kserve-self-signed-certs
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-csgnh
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:52Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:57Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:13:51Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:13:51Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:51Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://9e8a9668a0cad548adff08bed4e6c3869a02b192d77c3878b18e95e6a70e5357
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:10:57Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt/models
        name: kserve-provision-location
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-csgnh
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://979f05641f2dc45b7ef8603aab95939dbb45daad300b03963c004b12db57a5e7
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://979f05641f2dc45b7ef8603aab95939dbb45daad300b03963c004b12db57a5e7
          exitCode: 0
          finishedAt: "2026-07-02T22:10:56Z"
          reason: Completed
          startedAt: "2026-07-02T22:10:52Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-csgnh
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.49
    podIPs:
    - ip: 10.132.0.49
    qosClass: Burstable
    startTime: "2026-07-02T22:10:51Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.40/23"],"mac_address":"0a:58:0a:84:00:28","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.40/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.40"
            ],
            "mac": "0a:58:0a:84:00:28",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:41:10Z"
    generateName: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-cf76987fb-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-with-7ca60146
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: both
      pod-template-hash: cf76987fb
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:10Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"fc8a5d94-d6fb-4126-83aa-c882e0649f2a"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"CA_BUNDLE_CONFIGMAP_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"CA_BUNDLE_VOLUME_MOUNT_POINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/etc/ssl/custom-certs"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/mnt"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"cabundle-cert"}:
              .: {}
              f:configMap:
                .: {}
                f:defaultMode: {}
                f:name: {}
              f:name: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:41:10Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:11Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.40"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:42:41Z"
    name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-cf76987fb-h66t4
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-cf76987fb
      uid: fc8a5d94-d6fb-4126-83aa-c882e0649f2a
    resourceVersion: "31954"
    uid: 53a1a466-be87-4ca6-8d00-b7e94507407f
  spec:
    containers:
    - args:
      - --enable-lora
      - --lora-modules
      - '''{"name":"lora-adapter-1","path":"/mnt/lora/lora-adapter-1"}'''
      - '''{"name":"publishers/kserve-ci-e2e-test/models/lora-adapter-1","path":"/mnt/lora/lora-adapter-1"}'''
      command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" "publishers/kserve-ci-e2e-test/models/facebook/opt-125m" \
          --port 8000 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 10
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 2
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 1
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt
        name: kserve-provision-location
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-75ds9
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      - hf://edbeeching/opt-125m-lora
      - /mnt/lora/lora-adapter-1
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: CA_BUNDLE_CONFIGMAP_NAME
        value: odh-kserve-custom-ca-bundle
      - name: CA_BUNDLE_VOLUME_MOUNT_POINT
        value: /etc/ssl/custom-certs
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt
        name: kserve-provision-location
      - mountPath: /etc/ssl/custom-certs
        name: cabundle-cert
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-75ds9
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: default
    serviceAccountName: default
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - emptyDir: {}
      name: tmp-dir
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv3e414c2ba058a022dfd694dbcbac5b51-kserve-self-signed-certs
    - emptyDir: {}
      name: kserve-provision-location
    - configMap:
        defaultMode: 420
        name: odh-kserve-custom-ca-bundle
      name: cabundle-cert
    - name: kube-api-access-75ds9
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:11Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:15Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:42:41Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:42:41Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:10Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://fba964952db540ca16053f4ae3a421733c16109a0bdd59ee095313579f27121a
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:41:16Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt
        name: kserve-provision-location
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-75ds9
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://ce283b998bb5d5635b4ba6c2a48a69fa3c40c707f3641978d2a32f6eef5c7527
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://ce283b998bb5d5635b4ba6c2a48a69fa3c40c707f3641978d2a32f6eef5c7527
          exitCode: 0
          finishedAt: "2026-07-02T21:41:15Z"
          reason: Completed
          startedAt: "2026-07-02T21:41:11Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt
        name: kserve-provision-location
      - mountPath: /etc/ssl/custom-certs
        name: cabundle-cert
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-75ds9
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.40
    podIPs:
    - ip: 10.132.0.40
    qosClass: Burstable
    startTime: "2026-07-02T21:41:10Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      app.kubernetes.io/version: 0.9.0
      certificates.kserve.io/expiration-v2: "true"
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.133.0.48/23"],"mac_address":"0a:58:0a:85:00:30","gateway_ips":["10.133.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.133.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.133.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.133.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.133.0.1"}],"ip_address":"10.133.0.48/23","gateway_ip":"10.133.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.133.0.48"
            ],
            "mac": "0a:58:0a:85:00:30",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:41:10Z"
    generateName: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-router-scheduler-766bf6cdf7-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-router-scheduler
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-with-7ca60146
      app.kubernetes.io/part-of: llminferenceservice
      pod-template-hash: 766bf6cdf7
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-139-32
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:10Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:app.kubernetes.io/version: {}
            f:certificates.kserve.io/expiration-v2: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"7c1d7a77-f06d-4547-a500-4d1cd52dd290"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":5557,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9002,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9003,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/tmp/tokenizer"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"STORAGE_ALLOW_PATTERNS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tokenizer-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-tmp"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-uds"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:41:10Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:11Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.133.0.48"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:41:44Z"
    name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-router-schevwrxh
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-router-scheduler-766bf6cdf7
      uid: 7c1d7a77-f06d-4547-a500-4d1cd52dd290
    resourceVersion: "30954"
    uid: 54b86ac2-a846-4324-b950-4178d9f8a450
  spec:
    containers:
    - args:
      - --config-text
      - |
        apiVersion: inference.networking.x-k8s.io/v1alpha1
        kind: EndpointPickerConfig
        plugins:
        - type: single-profile-handler
        - type: queue-scorer
        - type: prefix-cache-scorer
        - type: max-score-picker
        - type: lora-affinity-scorer
        schedulingProfiles:
        - name: default
          plugins:
          - pluginRef: lora-affinity-scorer
            weight: 4
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
      command:
      - /app/epp
      - --pool-name
      - llmisvc-model-fb-opt-125m-with-7ca60146-inference-pool
      - --pool-namespace
      - kserve-ci-e2e-test
      - --zap-encoder
      - json
      - --grpc-port
      - "9002"
      - --grpc-health-port
      - "9003"
      - --enable-cert-reload=true
      - --secure-serving=true
      - --model-server-metrics-scheme=https
      - --cert-path=/var/run/kserve/tls
      env:
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: liveness
        initialDelaySeconds: 5
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 9002
        name: grpc
        protocol: TCP
      - containerPort: 9003
        name: grpc-health
        protocol: TCP
      - containerPort: 9090
        name: metrics
        protocol: TCP
      - containerPort: 5557
        name: zmq
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: readiness
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-hbzw2
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-fb-opt-125m-with-7ca60146-epp-sa-dockercfg-pnk5d
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: STORAGE_ALLOW_PATTERNS
        value: '["tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
          "vocab.json", "merges.txt", "config.json", "generation_config.json"]'
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-hbzw2
        readOnly: true
    nodeName: ip-10-0-139-32.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-fb-opt-125m-with-7ca60146-epp-sa
    serviceAccountName: llmisvc-model-fb-opt-125m-with-7ca60146-epp-sa
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv3e414c2ba058a022dfd694dbcbac5b51-kserve-self-signed-certs
    - emptyDir: {}
      name: tokenizer-uds
    - emptyDir: {}
      name: tokenizer-tmp
    - emptyDir: {}
      name: tokenizer-cache
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-hbzw2
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:11Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:12Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:44Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:44Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:41:10Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 256m
        memory: 500Mi
      containerID: cri-o://abadb50ad749398d5a25ae5de5382f2fb70df3a317b7d680ce918885f7152dcd
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-endpoint-picker@sha256:06b6c75d77afd0e07053402752a9736c2dfbc12a306d0d37d963aac4c1d4e6a6
      lastState: {}
      name: main
      ready: true
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:41:12Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-hbzw2
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.139.32
    hostIPs:
    - ip: 10.0.139.32
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://8d98763d54686103c1a43a5b6cc9f81bb77483c9dc464a9a20425e07db042a16
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://8d98763d54686103c1a43a5b6cc9f81bb77483c9dc464a9a20425e07db042a16
          exitCode: 0
          finishedAt: "2026-07-02T21:41:12Z"
          reason: Completed
          startedAt: "2026-07-02T21:41:11Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-hbzw2
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.133.0.48
    podIPs:
    - ip: 10.133.0.48
    qosClass: Burstable
    startTime: "2026-07-02T21:41:10Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.43/23"],"mac_address":"0a:58:0a:84:00:2b","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.43/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.43"
            ],
            "mac": "0a:58:0a:84:00:2b",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:57:59Z"
    generateName: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-598d446855-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-with-ba4d693a
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: both
      pod-template-hash: 598d446855
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T21:57:59Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"599993e1-aa92-4fc6-bc15-7aedfdda19b2"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"CA_BUNDLE_CONFIGMAP_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"CA_BUNDLE_VOLUME_MOUNT_POINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/etc/ssl/custom-certs"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/mnt"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"cabundle-cert"}:
              .: {}
              f:configMap:
                .: {}
                f:defaultMode: {}
                f:name: {}
              f:name: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:57:59Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:58:00Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.43"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:59:20Z"
    name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-598d446855-8lvfh
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-598d446855
      uid: 599993e1-aa92-4fc6-bc15-7aedfdda19b2
    resourceVersion: "49322"
    uid: af680406-b428-4781-a163-7131215ed8a1
  spec:
    containers:
    - args:
      - --enable-lora
      - --lora-modules
      - '''{"name":"lora-adapter-1","path":"/mnt/lora/lora-adapter-1"}'''
      - '''{"name":"publishers/kserve-ci-e2e-test/models/lora-adapter-1","path":"/mnt/lora/lora-adapter-1"}'''
      command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" "publishers/kserve-ci-e2e-test/models/facebook/opt-125m" \
          --port 8000 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 10
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 2
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 1
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt
        name: kserve-provision-location
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-qckfx
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      - hf://edbeeching/opt-125m-lora
      - /mnt/lora/lora-adapter-1
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: CA_BUNDLE_CONFIGMAP_NAME
        value: odh-kserve-custom-ca-bundle
      - name: CA_BUNDLE_VOLUME_MOUNT_POINT
        value: /etc/ssl/custom-certs
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt
        name: kserve-provision-location
      - mountPath: /etc/ssl/custom-certs
        name: cabundle-cert
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-qckfx
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: default
    serviceAccountName: default
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - emptyDir: {}
      name: tmp-dir
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv77ff2528d3e9b4972cd9335229fce9f0-kserve-self-signed-certs
    - emptyDir: {}
      name: kserve-provision-location
    - configMap:
        defaultMode: 420
        name: odh-kserve-custom-ca-bundle
      name: cabundle-cert
    - name: kube-api-access-qckfx
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:01Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:04Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:59:20Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:59:20Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:57:59Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://b2d451ea78726e6a6c4bfb4a01a26019d27dbdcfebecadced2f53ac8d64cb5dc
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:58:04Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt
        name: kserve-provision-location
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-qckfx
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://4c9d1ee751ccfbff144b07df27fcaad256e201f699e89d79d3ef65a4991d2a8f
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://4c9d1ee751ccfbff144b07df27fcaad256e201f699e89d79d3ef65a4991d2a8f
          exitCode: 0
          finishedAt: "2026-07-02T21:58:03Z"
          reason: Completed
          startedAt: "2026-07-02T21:58:00Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt
        name: kserve-provision-location
      - mountPath: /etc/ssl/custom-certs
        name: cabundle-cert
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-qckfx
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.43
    podIPs:
    - ip: 10.132.0.43
    qosClass: Burstable
    startTime: "2026-07-02T21:57:59Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      app.kubernetes.io/version: 0.9.0
      certificates.kserve.io/expiration-v2: "true"
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.133.0.51/23"],"mac_address":"0a:58:0a:85:00:33","gateway_ips":["10.133.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.133.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.133.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.133.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.133.0.1"}],"ip_address":"10.133.0.51/23","gateway_ip":"10.133.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.133.0.51"
            ],
            "mac": "0a:58:0a:85:00:33",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:57:59Z"
    generateName: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-router-scheduler-67cc76c46d-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-router-scheduler
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-with-ba4d693a
      app.kubernetes.io/part-of: llminferenceservice
      pod-template-hash: 67cc76c46d
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-139-32
      operation: Update
      subresource: status
      time: "2026-07-02T21:57:59Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:app.kubernetes.io/version: {}
            f:certificates.kserve.io/expiration-v2: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"68ac9d71-9583-42be-b18f-f409a2701638"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":5557,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9002,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9003,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/tmp/tokenizer"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"STORAGE_ALLOW_PATTERNS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tokenizer-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-tmp"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-uds"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:57:59Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:58:00Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.133.0.51"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:58:34Z"
    name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-router-schefgzsz
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-router-scheduler-67cc76c46d
      uid: 68ac9d71-9583-42be-b18f-f409a2701638
    resourceVersion: "48623"
    uid: 62b92b19-cbc6-44e7-9ce2-96af550e34b5
  spec:
    containers:
    - args:
      - --config-text
      - |
        apiVersion: inference.networking.x-k8s.io/v1alpha1
        kind: EndpointPickerConfig
        plugins:
        - type: single-profile-handler
        - type: queue-scorer
        - type: prefix-cache-scorer
        - type: max-score-picker
        - type: lora-affinity-scorer
        schedulingProfiles:
        - name: default
          plugins:
          - pluginRef: lora-affinity-scorer
            weight: 4
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
      command:
      - /app/epp
      - --pool-name
      - llmisvc-model-fb-opt-125m-with-ba4d693a-inference-pool
      - --pool-namespace
      - kserve-ci-e2e-test
      - --zap-encoder
      - json
      - --grpc-port
      - "9002"
      - --grpc-health-port
      - "9003"
      - --enable-cert-reload=true
      - --secure-serving=true
      - --model-server-metrics-scheme=https
      - --cert-path=/var/run/kserve/tls
      env:
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: liveness
        initialDelaySeconds: 5
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 9002
        name: grpc
        protocol: TCP
      - containerPort: 9003
        name: grpc-health
        protocol: TCP
      - containerPort: 9090
        name: metrics
        protocol: TCP
      - containerPort: 5557
        name: zmq
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: readiness
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-dwwvc
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-fb-opt-125m-with-ba4d693a-epp-sa-dockercfg-btgxm
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: STORAGE_ALLOW_PATTERNS
        value: '["tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
          "vocab.json", "merges.txt", "config.json", "generation_config.json"]'
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-dwwvc
        readOnly: true
    nodeName: ip-10-0-139-32.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-fb-opt-125m-with-ba4d693a-epp-sa
    serviceAccountName: llmisvc-model-fb-opt-125m-with-ba4d693a-epp-sa
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv77ff2528d3e9b4972cd9335229fce9f0-kserve-self-signed-certs
    - emptyDir: {}
      name: tokenizer-uds
    - emptyDir: {}
      name: tokenizer-tmp
    - emptyDir: {}
      name: tokenizer-cache
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-dwwvc
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:01Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:02Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:34Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:58:34Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:57:59Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 256m
        memory: 500Mi
      containerID: cri-o://e8dc01c0da869acb574ef91a213105d0c301c64b32de5adc2d580f3a1aaf2b90
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-endpoint-picker@sha256:06b6c75d77afd0e07053402752a9736c2dfbc12a306d0d37d963aac4c1d4e6a6
      lastState: {}
      name: main
      ready: true
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:58:02Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-dwwvc
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.139.32
    hostIPs:
    - ip: 10.0.139.32
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://3a4b2496bd70d15eb3d2b7e0fb2b77364476496b04c83a67c1930b2216c87ea9
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://3a4b2496bd70d15eb3d2b7e0fb2b77364476496b04c83a67c1930b2216c87ea9
          exitCode: 0
          finishedAt: "2026-07-02T21:58:01Z"
          reason: Completed
          startedAt: "2026-07-02T21:58:00Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-dwwvc
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.133.0.51
    podIPs:
    - ip: 10.133.0.51
    qosClass: Burstable
    startTime: "2026-07-02T21:57:59Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.46/23"],"mac_address":"0a:58:0a:84:00:2e","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.46/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.46"
            ],
            "mac": "0a:58:0a:84:00:2e",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:08:28Z"
    generateName: llmisvc-model-pvc-router-manage-e8706282-kserve-7db79954c-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload
      app.kubernetes.io/name: llmisvc-model-pvc-router-manage-e8706282
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: decode
      pod-template-hash: 7db79954c
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"cdb8615f-ebc3-4a82-a396-880b89a725e4"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8001,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"llm-d-routing-sidecar"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"INFERENCE_POOL_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"INFERENCE_POOL_NAMESPACE"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources: {}
              f:restartPolicy: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-pvc-source"}:
              .: {}
              f:name: {}
              f:persistentVolumeClaim:
                .: {}
                f:claimName: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:08:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:31Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.46"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:11:31Z"
    name: llmisvc-model-pvc-router-manage-e8706282-kserve-7db79954c-md987
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-pvc-router-manage-e8706282-kserve-7db79954c
      uid: cdb8615f-ebc3-4a82-a396-880b89a725e4
    resourceVersion: "60806"
    uid: f04ca879-43d8-4aea-a313-d3abe7c7557d
  spec:
    containers:
    - command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" "publishers/kserve-ci-e2e-test/models/facebook/opt-125m" \
          --port 8001 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 8
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 180
        periodSeconds: 30
        successThreshold: 1
        timeoutSeconds: 30
      name: main
      ports:
      - containerPort: 8001
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8001
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt/models
        name: kserve-pvc-source
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-2cmgg
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-pvc-router-manage-e8706282-kserve-dockercfg-rqjr7
    initContainers:
    - command:
      - /app/pd-sidecar
      - --port=8000
      - --vllm-port=8001
      - --kv-connector=nixlv2
      - --enable-ssrf-protection=true
      - --pool-group=inference.networking.x-k8s.io
      - --inference-pool=kserve-ci-e2e-test/llmisvc-model-pvc-router-manage-e8706282-inference-pool
      - --secure-proxy=true
      - --cert-path=/var/run/kserve/tls
      - --decoder-use-tls=true
      - --prefiller-use-tls=true
      env:
      - name: INFERENCE_POOL_NAMESPACE
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.namespace
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      - name: INFERENCE_POOL_NAME
        value: llmisvc-model-pvc-router-manage-e8706282-inference-pool
      image: ghcr.io/llm-d/llm-d-router-disagg-sidecar:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      livenessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 10
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 10
      name: llm-d-routing-sidecar
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 10
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 10
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources: {}
      restartPolicy: Always
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-2cmgg
        readOnly: true
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-pvc-router-manage-e8706282-kserve
    serviceAccountName: llmisvc-model-pvc-router-manage-e8706282-kserve
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir: {}
      name: tmp-dir
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisvb19f98874e050eec8ca94d49676113f0-kserve-self-signed-certs
    - name: kserve-pvc-source
      persistentVolumeClaim:
        claimName: e2e-pvc-model-storage
    - name: kube-api-access-2cmgg
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:33Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:33Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:31Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:31Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:28Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://cb1236233d12bbe0508f6b9541776db5a5e153dfa0caec96dad3bcb2dd1f069b
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:08:33Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt/models
        name: kserve-pvc-source
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-2cmgg
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    initContainerStatuses:
    - containerID: cri-o://e8b8d215bb1408634f7eb13f1740fb73a01d161afe9511107d116303b91eba66
      image: ghcr.io/llm-d/llm-d-router-disagg-sidecar:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-disagg-sidecar@sha256:1ec2c035216ea0a4f8a6f7ccda55398fc3d9f71f487c135f49ba0dea08163f12
      lastState: {}
      name: llm-d-routing-sidecar
      ready: true
      resources: {}
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:08:32Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-2cmgg
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.46
    podIPs:
    - ip: 10.132.0.46
    qosClass: Burstable
    startTime: "2026-07-02T22:08:28Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.47/23"],"mac_address":"0a:58:0a:84:00:2f","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.47/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.47"
            ],
            "mac": "0a:58:0a:84:00:2f",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:08:28Z"
    generateName: llmisvc-model-pvc-router-manage-e8706282-kserve-prefill-79497db4cc-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-workload-prefill
      app.kubernetes.io/name: llmisvc-model-pvc-router-manage-e8706282
      app.kubernetes.io/part-of: llminferenceservice
      kserve.io/component: workload
      llm-d.ai/role: prefill
      pod-template-hash: 79497db4cc
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:kserve.io/component: {}
            f:llm-d.ai/role: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"0a559d1b-3a24-4322-8a1d-955ff0896cd2"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"HF_HUB_CACHE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HOME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"TORCHINDUCTOR_CACHE_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"USER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_CPU_KVCACHE_SPACE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_ENABLE_V1_MULTIPROCESSING"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"VLLM_LOGGING_LEVEL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":8000,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/dev/shm"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/home"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
                k:{"mountPath":"/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/tmp"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"dshm"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
                f:sizeLimit: {}
              f:name: {}
            k:{"name":"home"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"kserve-pvc-source"}:
              .: {}
              f:name: {}
              f:persistentVolumeClaim:
                .: {}
                f:claimName: {}
            k:{"name":"model-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tmp-dir"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:08:28Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:31Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.47"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:11:31Z"
    name: llmisvc-model-pvc-router-manage-e8706282-kserve-prefill-79zlqf4
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc-model-pvc-router-manage-e8706282-kserve-prefill-79497db4cc
      uid: 0a559d1b-3a24-4322-8a1d-955ff0896cd2
    resourceVersion: "60816"
    uid: ecb4a255-8928-4569-8b05-c860fa5adfd6
  spec:
    containers:
    - command:
      - /bin/bash
      - -c
      - |-
        if [ -f /etc/profile.d/ibm-aiu-setup.sh ]; then
          source /etc/profile.d/ibm-aiu-setup.sh
        fi

        if [ "$KSERVE_INFER_ROCE" = "true" ]; then
          echo "Trying to infer RoCE configs ... "
          grep -H . /sys/class/infiniband/*/ports/*/gids/* 2>/dev/null
          grep -H . /sys/class/infiniband/*/ports/*/gid_attrs/types/* 2>/dev/null

          cat /proc/driver/nvidia/params

          KSERVE_INFER_IB_GID_INDEX_GREP=${KSERVE_INFER_IB_GID_INDEX_GREP:-"RoCE v2"}

          echo "[Infer RoCE] Discovering active HCAs ..."
          active_hcas=()
          # Loop through all mlx5 devices found in sysfs
          for hca_dir in /sys/class/infiniband/mlx5_*; do
              # Ensure it's a directory before proceeding
              if [ -d "$hca_dir" ]; then
                  hca_name=$(basename "$hca_dir")
                  port_state_file="$hca_dir/ports/1/state" # Assume port 1
                  type_file="$hca_dir/ports/1/gid_attrs/types/*"

                  echo "[Infer RoCE] Check if the port state file ${port_state_file} exists and contains 'ACTIVE'"
                  if [ -f "$port_state_file" ] && grep -q "ACTIVE" "$port_state_file" && grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" ${type_file} 2>/dev/null; then
                      echo "[Infer RoCE] Found active HCA: $hca_name"
                      active_hcas+=("$hca_name")
                  else
                      echo "[Infer RoCE] Skipping inactive or down HCA: $hca_name"
                  fi
              fi
          done

          # Check if we found any active HCAs
          if [ ${#active_hcas[@]} -gt 0 ]; then
              # Join the array elements with a comma
              hca_port_pairs=()
              for hca in "${active_hcas[@]}"; do
                hca_port_pairs+=("${hca}:1")
              done

              active_hca_list=$(IFS=,; echo "${active_hcas[*]}")
              hca_port_pairs_list=$(IFS=,; echo "${hca_port_pairs[*]}")
              echo "[Infer RoCE] Setting active HCAs: ${active_hca_list}"
              export NCCL_IB_HCA=${NCCL_IB_HCA:-${active_hca_list}}
              export NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST:-${hca_port_pairs_list}}
              export UCX_NET_DEVICES=${UCX_NET_DEVICES:-${hca_port_pairs_list}}

              echo "[Infer RoCE] NCCL_IB_HCA=${NCCL_IB_HCA}"
              echo "[Infer RoCE] NVSHMEM_HCA_LIST=${NVSHMEM_HCA_LIST}"
              echo "[Infer RoCE] UCX_NET_DEVICES=${UCX_NET_DEVICES}"
          else
              echo "[Infer RoCE] WARNING: No active RoCE HCAs found. NCCL_IB_HCA will not be set."
          fi

          if [ ${#active_hcas[@]} -gt 0 ]; then
              echo "[Infer RoCE] Finding GID_INDEX for each active HCA (SR-IOV compatible)..."

              # For SR-IOV environments, find the most common IPv4 RoCE v2 GID index across all HCAs
              declare -A gid_index_count
              declare -A hca_gid_index

              for hca_name in "${active_hcas[@]}"; do
                  echo "[Infer RoCE] Processing HCA: ${hca_name}"

                  # Find all RoCE v2 IPv4 GIDs for this HCA and count by index
                  for tpath in /sys/class/infiniband/${hca_name}/ports/1/gid_attrs/types/*; do
                      if grep -q "${KSERVE_INFER_IB_GID_INDEX_GREP}" "$tpath" 2>/dev/null; then
                          idx=$(basename "$tpath")
                          gid_file="/sys/class/infiniband/${hca_name}/ports/1/gids/${idx}"
                          # Check for IPv4 GID (contains ffff:)
                          if [ -f "$gid_file" ] && grep -q "ffff:" "$gid_file"; then
                              gid_value=$(cat "$gid_file" 2>/dev/null || echo "")
                              echo "[Infer RoCE] Found IPv4 RoCE v2 GID for ${hca_name}: index=${idx}, gid=${gid_value}"
                              hca_gid_index["${hca_name}"]="${idx}"
                              gid_index_count["${idx}"]=$((${gid_index_count["${idx}"]} + 1))
                              break  # Use first found IPv4 GID per HCA
                          fi
                      fi
                  done
              done

              # Find the most common GID index (most likely to be consistent across nodes)
              best_gid_index=""
              max_count=0
              for idx in "${!gid_index_count[@]}"; do
                  count=${gid_index_count["${idx}"]}
                  echo "[Infer RoCE] GID_INDEX ${idx} found on ${count} HCAs"
                  if [ $count -gt $max_count ]; then
                      max_count=$count
                      best_gid_index="$idx"
                  fi
              done

              # Use deterministic fallback if tied - prefer index 3 (SR-IOV standard)
              if [ ${#gid_index_count[@]} -gt 1 ]; then
                  echo "[Infer RoCE] Multiple GID indices found, selecting most common: ${best_gid_index}"
                  # If there's a tie, prefer index 3 as it's most common in SR-IOV setups
                  if [ -n "${gid_index_count['3']}" ] && [ "${gid_index_count['3']}" -eq "$max_count" ]; then
                      best_gid_index="3"
                      echo "[Infer RoCE] Using deterministic fallback: GID_INDEX=3 (SR-IOV standard)"
                  fi
              fi

              # Check if GID_INDEX is already set via environment variables
              if [ -n "${NCCL_IB_GID_INDEX}" ]; then
                  echo "[Infer RoCE] Using pre-configured NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX} from environment"
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$NCCL_IB_GID_INDEX}
                  echo "[Infer RoCE] Using pre-configured GID_INDEX=${NCCL_IB_GID_INDEX} for NCCL, NVSHMEM, and UCX"
              elif [ -n "$best_gid_index" ]; then
                  echo "[Infer RoCE] Selected GID_INDEX: ${best_gid_index} (found on ${max_count} HCAs)"

                  export NCCL_IB_GID_INDEX=${NCCL_IB_GID_INDEX:-$best_gid_index}
                  export NVSHMEM_IB_GID_INDEX=${NVSHMEM_IB_GID_INDEX:-$best_gid_index}
                  export UCX_IB_GID_INDEX=${UCX_IB_GID_INDEX:-$best_gid_index}

                  echo "[Infer RoCE] Exported GID_INDEX=${best_gid_index} for NCCL, NVSHMEM, and UCX"
              else
                  echo "[Infer RoCE] ERROR: No valid IPv4 ${KSERVE_INFER_IB_GID_INDEX_GREP} GID_INDEX found on any HCA."
              fi
          else
              echo "[Infer RoCE] No active HCAs found, skipping GID_INDEX inference."
          fi
        fi

        # --disable-access-log-for-endpoints landed in vLLM 0.16.0 (vllm-project/vllm#30011).
        # Older versions still need the blanket --disable-uvicorn-access-log.
        ACCESS_LOG_ARGS="--disable-uvicorn-access-log"
        VLLM_VERSION=$(vllm --version 2>/dev/null | tail -1 | awk '{print $NF}')
        echo "[access-log-detect] vllm version='${VLLM_VERSION}'"
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.16.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.16.0" ]; then
          ACCESS_LOG_ARGS="--disable-access-log-for-endpoints /health,/metrics,/ping"
        fi
        echo "[access-log-detect] selected ACCESS_LOG_ARGS='${ACCESS_LOG_ARGS}'"

        # --shutdown-timeout landed in vLLM 0.18.0 (vllm-project/vllm#36666).
        SHUTDOWN_TIMEOUT_ARGS=""
        if [[ "$VLLM_VERSION" =~ ^[0-9]+\.[0-9]+ ]] && [ "$(printf '%s\n%s\n' "0.18.0" "${VLLM_VERSION}" | sort -V | head -1)" = "0.18.0" ]; then
          SHUTDOWN_TIMEOUT_ARGS="--shutdown-timeout 40"
        fi

        eval "exec vllm serve /mnt/models \
          --served-model-name "facebook/opt-125m" \
          --port 8000 \
          ${ACCESS_LOG_ARGS} \
          ${SHUTDOWN_TIMEOUT_ARGS} \
          --enable-ssl-refresh \
          --ssl-certfile /var/run/kserve/tls/tls.crt \
          --ssl-keyfile /var/run/kserve/tls/tls.key \
          ${VLLM_ADDITIONAL_ARGS} \
          $@"
      - --
      env:
      - name: HOME
        value: /home
      - name: VLLM_LOGGING_LEVEL
        value: DEBUG
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_ENABLE_V1_MULTIPROCESSING
        value: "0"
      - name: USER
        value: nonroot
      - name: TORCHINDUCTOR_CACHE_DIR
        value: /tmp/torchinductor-cache
      - name: HF_HUB_CACHE
        value: /models
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 8
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 180
        periodSeconds: 30
        successThreshold: 1
        timeoutSeconds: 30
      name: main
      ports:
      - containerPort: 8000
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 5
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      startupProbe:
        failureThreshold: 60
        httpGet:
          path: /health
          port: 8000
          scheme: HTTPS
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: File
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /mnt/models
        name: kserve-pvc-source
        readOnly: true
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-4r9rp
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: default
    serviceAccountName: default
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: home
    - emptyDir: {}
      name: tmp-dir
    - emptyDir:
        medium: Memory
        sizeLimit: 1Gi
      name: dshm
    - emptyDir: {}
      name: model-cache
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisvb19f98874e050eec8ca94d49676113f0-kserve-self-signed-certs
    - name: kserve-pvc-source
      persistentVolumeClaim:
        claimName: e2e-pvc-model-storage
    - name: kube-api-access-4r9rp
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:32Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:28Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:31Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:31Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:28Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 200m
        memory: 2Gi
      containerID: cri-o://7bec354571f1bd3f1be21b9aec69291a0075eedebe5cd5b0ccca2e266953cad0
      image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0
      imageID: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo@sha256:afb39fca138b51d019d986229d546531b45a2a3deb73bcf59bd42406e13fbba0
      lastState: {}
      name: main
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 7Gi
        requests:
          cpu: 200m
          memory: 2Gi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:08:31Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /home
        name: home
      - mountPath: /tmp
        name: tmp-dir
      - mountPath: /dev/shm
        name: dshm
      - mountPath: /models
        name: model-cache
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /mnt/models
        name: kserve-pvc-source
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-4r9rp
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.47
    podIPs:
    - ip: 10.132.0.47
    qosClass: Burstable
    startTime: "2026-07-02T22:08:28Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      app.kubernetes.io/version: 0.9.0
      certificates.kserve.io/expiration-v2: "true"
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.133.0.54/23"],"mac_address":"0a:58:0a:85:00:36","gateway_ips":["10.133.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.133.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.133.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.133.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.133.0.1"}],"ip_address":"10.133.0.54/23","gateway_ip":"10.133.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.133.0.54"
            ],
            "mac": "0a:58:0a:85:00:36",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:10:51Z"
    generateName: llmisvc44d181485fad85e662eb092f3749502f-kserve-router-scheduler-69454684c7-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-router-scheduler
      app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-50bc673d
      app.kubernetes.io/part-of: llminferenceservice
      pod-template-hash: 69454684c7
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-139-32
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:app.kubernetes.io/version: {}
            f:certificates.kserve.io/expiration-v2: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"1f3a1682-99d0-41fe-9dc7-a09b556b2cd4"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":5557,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9002,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9003,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/tmp/tokenizer"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:initContainers:
            .: {}
            k:{"name":"storage-initializer"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"AWS_ACCESS_KEY_ID"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"AWS_CA_BUNDLE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_CA_BUNDLE_CONFIGMAP"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_ENDPOINT_URL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"AWS_SECRET_ACCESS_KEY"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:secretKeyRef: {}
                k:{"name":"HF_HUB_ENABLE_HF_TRANSFER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_HIGH_PERFORMANCE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"HF_XET_NUM_CONCURRENT_RANGE_GETS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_ENDPOINT"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_USE_HTTPS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"S3_VERIFY_SSL"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"STORAGE_ALLOW_PATTERNS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/mnt/models"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"kserve-provision-location"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tokenizer-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-tmp"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-uds"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:10:51Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:10:52Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:initContainerStatuses: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.133.0.54"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:11:25Z"
    name: llmisvc44d181485fad85e662eb092f3749502f-kserve-router-schepcq9n
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvc44d181485fad85e662eb092f3749502f-kserve-router-scheduler-69454684c7
      uid: 1f3a1682-99d0-41fe-9dc7-a09b556b2cd4
    resourceVersion: "60713"
    uid: c5c7d80e-0dfa-46ea-b141-8d34db09e58e
  spec:
    containers:
    - args:
      - --config-text
      - |
        apiVersion: inference.networking.x-k8s.io/v1alpha1
        kind: EndpointPickerConfig
        plugins:
        - type: disagg-headers-handler
        - type: prefill-filter
        - type: decode-filter
        - type: queue-scorer
        - type: prefix-cache-scorer
        - type: max-score-picker
        - type: always-disagg-pd-decider
        - parameters:
            deciders:
              prefill: always-disagg-pd-decider
          type: disagg-profile-handler
        schedulingProfiles:
        - name: prefill
          plugins:
          - pluginRef: prefill-filter
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
        - name: decode
          plugins:
          - pluginRef: decode-filter
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
      command:
      - /app/epp
      - --pool-name
      - llmisvc-model-fb-opt-125m-route-50bc673d-inference-pool
      - --pool-namespace
      - kserve-ci-e2e-test
      - --zap-encoder
      - json
      - --grpc-port
      - "9002"
      - --grpc-health-port
      - "9003"
      - --enable-cert-reload=true
      - --secure-serving=true
      - --model-server-metrics-scheme=https
      - --cert-path=/var/run/kserve/tls
      env:
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: liveness
        initialDelaySeconds: 5
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 9002
        name: grpc
        protocol: TCP
      - containerPort: 9003
        name: grpc-health
        protocol: TCP
      - containerPort: 9090
        name: metrics
        protocol: TCP
      - containerPort: 5557
        name: zmq
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: readiness
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-n5f9s
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-fb-opt-125m-route-50bc673d-epp-sa-dockercfg-mj95w
    initContainers:
    - args:
      - hf://facebook/opt-125m
      - /mnt/models
      env:
      - name: AWS_ACCESS_KEY_ID
        valueFrom:
          secretKeyRef:
            key: AWS_ACCESS_KEY_ID
            name: seaweedfs-s3-creds
      - name: AWS_SECRET_ACCESS_KEY
        valueFrom:
          secretKeyRef:
            key: AWS_SECRET_ACCESS_KEY
            name: seaweedfs-s3-creds
      - name: S3_USE_HTTPS
        value: "0"
      - name: S3_ENDPOINT
        value: s3-service.kserve:8333
      - name: AWS_ENDPOINT_URL
        value: http://s3-service.kserve:8333
      - name: S3_VERIFY_SSL
        value: "0"
      - name: AWS_CA_BUNDLE
        value: /etc/ssl/custom-certs/cabundle.crt
      - name: AWS_CA_BUNDLE_CONFIGMAP
        value: odh-kserve-custom-ca-bundle
      - name: HF_HUB_ENABLE_HF_TRANSFER
        value: "1"
      - name: HF_XET_HIGH_PERFORMANCE
        value: "1"
      - name: HF_XET_NUM_CONCURRENT_RANGE_GETS
        value: "8"
      - name: STORAGE_ALLOW_PATTERNS
        value: '["tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
          "vocab.json", "merges.txt", "config.json", "generation_config.json"]'
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imagePullPolicy: IfNotPresent
      name: storage-initializer
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        runAsNonRoot: true
        runAsUser: 1000690000
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-n5f9s
        readOnly: true
    nodeName: ip-10-0-139-32.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-fb-opt-125m-route-50bc673d-epp-sa
    serviceAccountName: llmisvc-model-fb-opt-125m-route-50bc673d-epp-sa
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisv44d181485fad85e662eb092f3749502f-kserve-self-signed-certs
    - emptyDir: {}
      name: tokenizer-uds
    - emptyDir: {}
      name: tokenizer-tmp
    - emptyDir: {}
      name: tokenizer-cache
    - emptyDir: {}
      name: kserve-provision-location
    - name: kube-api-access-n5f9s
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:53Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:53Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:25Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:11:25Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:10:51Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 256m
        memory: 500Mi
      containerID: cri-o://c14fd4ce889168a86220cc88164ff39802274cd0b1a6616d8590b8f52503a48b
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-endpoint-picker@sha256:06b6c75d77afd0e07053402752a9736c2dfbc12a306d0d37d963aac4c1d4e6a6
      lastState: {}
      name: main
      ready: true
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:10:53Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-n5f9s
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.139.32
    hostIPs:
    - ip: 10.0.139.32
    initContainerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 100Mi
      containerID: cri-o://dafc7b7d5ea0455d15cd90900286248a72454ef58d401035c088b8c1edbc9744
      image: quay.io/opendatahub/kserve-storage-initializer@sha256:33f82c067d520fe749f73749cf76b5935fe69bd9eb141504c6660698c7947743
      imageID: quay.io/opendatahub/kserve-storage-initializer@sha256:08453dfdb92f739bac773ed802df8db34a80ade6399c0e69ec169c60d78d911d
      lastState: {}
      name: storage-initializer
      ready: true
      resources:
        limits:
          cpu: "1"
          memory: 24Gi
        requests:
          cpu: 100m
          memory: 100Mi
      restartCount: 0
      started: false
      state:
        terminated:
          containerID: cri-o://dafc7b7d5ea0455d15cd90900286248a72454ef58d401035c088b8c1edbc9744
          exitCode: 0
          finishedAt: "2026-07-02T22:10:53Z"
          reason: Completed
          startedAt: "2026-07-02T22:10:52Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /mnt/models
        name: kserve-provision-location
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-n5f9s
        readOnly: true
        recursiveReadOnly: Disabled
    observedGeneration: 1
    phase: Running
    podIP: 10.133.0.54
    podIPs:
    - ip: 10.133.0.54
    qosClass: Burstable
    startTime: "2026-07-02T22:10:51Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      app.kubernetes.io/version: 0.9.0
      certificates.kserve.io/expiration-v2: "true"
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.134.0.35/23"],"mac_address":"0a:58:0a:86:00:23","gateway_ips":["10.134.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.134.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.134.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.134.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.134.0.1"}],"ip_address":"10.134.0.35/23","gateway_ip":"10.134.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.134.0.35"
            ],
            "mac": "0a:58:0a:86:00:23",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:08:29Z"
    generateName: llmisvcb19f98874e050eec8ca94d49676113f0-kserve-router-scheduler-67b4bb9646-
    generation: 1
    labels:
      app.kubernetes.io/component: llminferenceservice-router-scheduler
      app.kubernetes.io/name: llmisvc-model-pvc-router-manage-e8706282
      app.kubernetes.io/part-of: llminferenceservice
      pod-template-hash: 67b4bb9646
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-139-243
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:29Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:app.kubernetes.io/version: {}
            f:certificates.kserve.io/expiration-v2: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:app.kubernetes.io/component: {}
            f:app.kubernetes.io/name: {}
            f:app.kubernetes.io/part-of: {}
            f:pod-template-hash: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"389b7fd3-13c7-4071-a820-cee61cc507fa"}: {}
        f:spec:
          f:containers:
            k:{"name":"main"}:
              .: {}
              f:args: {}
              f:command: {}
              f:env:
                .: {}
                k:{"name":"SSL_CERT_DIR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:lifecycle:
                .: {}
                f:preStop:
                  .: {}
                  f:exec:
                    .: {}
                    f:command: {}
              f:livenessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":5557,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9002,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9003,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":9090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:grpc:
                  .: {}
                  f:port: {}
                  f:service: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:readOnlyRootFilesystem: {}
                f:runAsNonRoot: {}
                f:seccompProfile:
                  .: {}
                  f:type: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/tmp/tokenizer"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/kserve/tls"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                  f:readOnly: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"kserve-pvc-source"}:
              .: {}
              f:name: {}
              f:persistentVolumeClaim:
                .: {}
                f:claimName: {}
            k:{"name":"tls-certs"}:
              .: {}
              f:name: {}
              f:secret:
                .: {}
                f:defaultMode: {}
                f:secretName: {}
            k:{"name":"tokenizer-cache"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-tmp"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"tokenizer-uds"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:08:29Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:08:29Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.134.0.35"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:09:00Z"
    name: llmisvcb19f98874e050eec8ca94d49676113f0-kserve-router-schejwtjs
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: llmisvcb19f98874e050eec8ca94d49676113f0-kserve-router-scheduler-67b4bb9646
      uid: 389b7fd3-13c7-4071-a820-cee61cc507fa
    resourceVersion: "58375"
    uid: 8ff7430f-9986-4aec-9126-06470f4cb787
  spec:
    containers:
    - args:
      - --config-text
      - |
        apiVersion: inference.networking.x-k8s.io/v1alpha1
        kind: EndpointPickerConfig
        plugins:
        - type: disagg-headers-handler
        - type: prefill-filter
        - type: decode-filter
        - type: queue-scorer
        - type: prefix-cache-scorer
        - type: max-score-picker
        - type: always-disagg-pd-decider
        - parameters:
            deciders:
              prefill: always-disagg-pd-decider
          type: disagg-profile-handler
        schedulingProfiles:
        - name: prefill
          plugins:
          - pluginRef: prefill-filter
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
        - name: decode
          plugins:
          - pluginRef: decode-filter
          - pluginRef: queue-scorer
            weight: 2
          - pluginRef: prefix-cache-scorer
            weight: 3
          - pluginRef: max-score-picker
      command:
      - /app/epp
      - --pool-name
      - llmisvc-model-pvc-router-manage-e8706282-inference-pool
      - --pool-namespace
      - kserve-ci-e2e-test
      - --zap-encoder
      - json
      - --grpc-port
      - "9002"
      - --grpc-health-port
      - "9003"
      - --enable-cert-reload=true
      - --secure-serving=true
      - --model-server-metrics-scheme=https
      - --cert-path=/var/run/kserve/tls
      env:
      - name: SSL_CERT_DIR
        value: /var/run/kserve/tls:/var/run/secrets/kubernetes.io/serviceaccount:/etc/pki/tls/certs
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imagePullPolicy: IfNotPresent
      lifecycle:
        preStop:
          exec:
            command:
            - /bin/sleep
            - "15"
      livenessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: liveness
        initialDelaySeconds: 5
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      name: main
      ports:
      - containerPort: 9002
        name: grpc
        protocol: TCP
      - containerPort: 9003
        name: grpc-health
        protocol: TCP
      - containerPort: 9090
        name: metrics
        protocol: TCP
      - containerPort: 5557
        name: zmq
        protocol: TCP
      readinessProbe:
        failureThreshold: 3
        grpc:
          port: 9003
          service: readiness
        initialDelaySeconds: 30
        periodSeconds: 10
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 1000690000
        seccompProfile:
          type: RuntimeDefault
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: FallbackToLogsOnError
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mpcs6
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: default-dockercfg-2bqkp
    - name: llmisvc-model-pvc-router-manage-e8706282-epp-sa-dockercfg-8qk9q
    nodeName: ip-10-0-139-243.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
    serviceAccount: llmisvc-model-pvc-router-manage-e8706282-epp-sa
    serviceAccountName: llmisvc-model-pvc-router-manage-e8706282-epp-sa
    terminationGracePeriodSeconds: 60
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - name: tls-certs
      secret:
        defaultMode: 420
        secretName: llmisvb19f98874e050eec8ca94d49676113f0-kserve-self-signed-certs
    - emptyDir: {}
      name: tokenizer-uds
    - emptyDir: {}
      name: tokenizer-tmp
    - emptyDir: {}
      name: tokenizer-cache
    - name: kserve-pvc-source
      persistentVolumeClaim:
        claimName: e2e-pvc-model-storage
    - name: kube-api-access-mpcs6
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:29Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:29Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:09:00Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:09:00Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:08:29Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 256m
        memory: 500Mi
      containerID: cri-o://b6df6ffe29e96c19b2d24dee45556c4a91bd83ee4ed3cfe799a68f513659c042
      image: ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0-rc.2
      imageID: ghcr.io/llm-d/llm-d-router-endpoint-picker@sha256:06b6c75d77afd0e07053402752a9736c2dfbc12a306d0d37d963aac4c1d4e6a6
      lastState: {}
      name: main
      ready: true
      resources:
        requests:
          cpu: 256m
          memory: 500Mi
      restartCount: 0
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:08:29Z"
      user:
        linux:
          gid: 0
          supplementalGroups:
          - 0
          - 1000690000
          uid: 1000690000
      volumeMounts:
      - mountPath: /var/run/kserve/tls
        name: tls-certs
        readOnly: true
        recursiveReadOnly: Disabled
      - mountPath: /tmp/tokenizer
        name: tokenizer-uds
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-mpcs6
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.139.243
    hostIPs:
    - ip: 10.0.139.243
    observedGeneration: 1
    phase: Running
    podIP: 10.134.0.35
    podIPs:
    - ip: 10.134.0.35
    qosClass: Burstable
    startTime: "2026-07-02T22:08:29Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      istio.io/rev: openshift-gateway
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.132.0.31/23"],"mac_address":"0a:58:0a:84:00:1f","gateway_ips":["10.132.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.132.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.132.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.132.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.132.0.1"}],"ip_address":"10.132.0.31/23","gateway_ip":"10.132.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.132.0.31"
            ],
            "mac": "0a:58:0a:84:00:1f",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      prometheus.io/path: /stats/prometheus
      prometheus.io/port: "15020"
      prometheus.io/scrape: "true"
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T21:31:26Z"
    generateName: router-gateway-1-openshift-default-799f46c59b-
    generation: 1
    labels:
      gateway.istio.io/managed: istio.io-gateway-controller
      gateway.networking.k8s.io/gateway-name: router-gateway-1
      pod-template-hash: 799f46c59b
      service.istio.io/canonical-name: router-gateway-1-openshift-default
      service.istio.io/canonical-revision: latest
      sidecar.istio.io/inject: "false"
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-130-231
      operation: Update
      subresource: status
      time: "2026-07-02T21:31:26Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:istio.io/rev: {}
            f:prometheus.io/path: {}
            f:prometheus.io/port: {}
            f:prometheus.io/scrape: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:gateway.istio.io/managed: {}
            f:gateway.networking.k8s.io/gateway-name: {}
            f:pod-template-hash: {}
            f:service.istio.io/canonical-name: {}
            f:service.istio.io/canonical-revision: {}
            f:sidecar.istio.io/inject: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"d4b299be-a9cd-41e1-808c-b6b5e4b0dc04"}: {}
        f:spec:
          f:containers:
            k:{"name":"istio-proxy"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"CA_ADDR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"GOMAXPROCS"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"GOMEMLIMIT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"HOST_IP"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"INSTANCE_IP"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"ISTIO_CPU_LIMIT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"ISTIO_META_APP_CONTAINERS"}:
                  .: {}
                  f:name: {}
                k:{"name":"ISTIO_META_CLUSTER_ID"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_INTERCEPTION_MODE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_MESH_ID"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_NODE_NAME"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"ISTIO_META_OWNER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_POD_PORTS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_WORKLOAD_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"PILOT_CERT_PROVIDER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"POD_NAME"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"POD_NAMESPACE"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"PROXY_CONFIG"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"SERVICE_ACCOUNT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"TRUST_DOMAIN"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":15020,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":15021,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":15090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:privileged: {}
                f:readOnlyRootFilesystem: {}
                f:runAsGroup: {}
                f:runAsNonRoot: {}
                f:runAsUser: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/etc/istio/pod"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/etc/istio/proxy"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/lib/istio/data"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/credential-uds"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/istio"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/tokens"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/workload-spiffe-credentials"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/workload-spiffe-uds"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext:
            .: {}
            f:sysctls: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"credential-socket"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"istio-data"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"istio-envoy"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
              f:name: {}
            k:{"name":"istio-podinfo"}:
              .: {}
              f:downwardAPI:
                .: {}
                f:defaultMode: {}
                f:items: {}
              f:name: {}
            k:{"name":"istio-token"}:
              .: {}
              f:name: {}
              f:projected:
                .: {}
                f:defaultMode: {}
                f:sources: {}
            k:{"name":"istiod-ca-cert"}:
              .: {}
              f:configMap:
                .: {}
                f:defaultMode: {}
                f:name: {}
              f:name: {}
            k:{"name":"workload-certs"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"workload-socket"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T21:31:26Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T21:31:27Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.132.0.31"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T21:31:51Z"
    name: router-gateway-1-openshift-default-799f46c59b-z2jmj
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: router-gateway-1-openshift-default-799f46c59b
      uid: d4b299be-a9cd-41e1-808c-b6b5e4b0dc04
    resourceVersion: "21958"
    uid: 1777c13a-a427-4f15-a425-9bf364c91344
  spec:
    containers:
    - args:
      - proxy
      - router
      - --domain
      - $(POD_NAMESPACE).svc.cluster.local
      - --proxyLogLevel
      - warning
      - --proxyComponentLogLevel
      - misc:error
      - --log_output_level
      - default:info
      env:
      - name: PILOT_CERT_PROVIDER
        value: istiod
      - name: CA_ADDR
        value: istiod-openshift-gateway.openshift-ingress.svc:15012
      - name: POD_NAME
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.name
      - name: POD_NAMESPACE
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.namespace
      - name: INSTANCE_IP
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: status.podIP
      - name: SERVICE_ACCOUNT
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: spec.serviceAccountName
      - name: HOST_IP
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: status.hostIP
      - name: ISTIO_CPU_LIMIT
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.cpu
      - name: PROXY_CONFIG
        value: |
          {"discoveryAddress":"istiod-openshift-gateway.openshift-ingress.svc:15012","proxyHeaders":{"server":{"disabled":true},"envoyDebugHeaders":{"disabled":true},"metadataExchangeHeaders":{"mode":"IN_MESH"}}}
      - name: ISTIO_META_POD_PORTS
        value: '[]'
      - name: ISTIO_META_APP_CONTAINERS
      - name: GOMEMLIMIT
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.memory
      - name: GOMAXPROCS
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.cpu
      - name: ISTIO_META_CLUSTER_ID
        value: Kubernetes
      - name: ISTIO_META_NODE_NAME
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: spec.nodeName
      - name: ISTIO_META_INTERCEPTION_MODE
        value: REDIRECT
      - name: ISTIO_META_WORKLOAD_NAME
        value: router-gateway-1-openshift-default
      - name: ISTIO_META_OWNER
        value: kubernetes://apis/apps/v1/namespaces/kserve-ci-e2e-test/deployments/router-gateway-1-openshift-default
      - name: ISTIO_META_MESH_ID
        value: cluster.local
      - name: TRUST_DOMAIN
        value: cluster.local
      image: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:7d15cebf9b62f3f235c0eab5158ac8ff2fda86a1d193490dc94c301402c99da8
      imagePullPolicy: IfNotPresent
      name: istio-proxy
      ports:
      - containerPort: 15020
        name: metrics
        protocol: TCP
      - containerPort: 15021
        name: status-port
        protocol: TCP
      - containerPort: 15090
        name: http-envoy-prom
        protocol: TCP
      readinessProbe:
        failureThreshold: 4
        httpGet:
          path: /healthz/ready
          port: 15021
          scheme: HTTP
        periodSeconds: 15
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: 100m
          memory: 128Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        privileged: false
        readOnlyRootFilesystem: true
        runAsGroup: 1000699999
        runAsNonRoot: true
        runAsUser: 1000699999
      startupProbe:
        failureThreshold: 30
        httpGet:
          path: /healthz/ready
          port: 15021
          scheme: HTTP
        initialDelaySeconds: 1
        periodSeconds: 1
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: File
      volumeMounts:
      - mountPath: /var/run/secrets/workload-spiffe-uds
        name: workload-socket
      - mountPath: /var/run/secrets/credential-uds
        name: credential-socket
      - mountPath: /var/run/secrets/workload-spiffe-credentials
        name: workload-certs
      - mountPath: /var/run/secrets/istio
        name: istiod-ca-cert
      - mountPath: /var/lib/istio/data
        name: istio-data
      - mountPath: /etc/istio/proxy
        name: istio-envoy
      - mountPath: /var/run/secrets/tokens
        name: istio-token
      - mountPath: /etc/istio/pod
        name: istio-podinfo
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-fp5jr
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: router-gateway-1-openshift-default-dockercfg-rhhvk
    nodeName: ip-10-0-130-231.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
      sysctls:
      - name: net.ipv4.ip_unprivileged_port_start
        value: "0"
    serviceAccount: router-gateway-1-openshift-default
    serviceAccountName: router-gateway-1-openshift-default
    terminationGracePeriodSeconds: 30
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: workload-socket
    - emptyDir: {}
      name: credential-socket
    - emptyDir: {}
      name: workload-certs
    - emptyDir:
        medium: Memory
      name: istio-envoy
    - emptyDir: {}
      name: istio-data
    - downwardAPI:
        defaultMode: 420
        items:
        - fieldRef:
            apiVersion: v1
            fieldPath: metadata.labels
          path: labels
        - fieldRef:
            apiVersion: v1
            fieldPath: metadata.annotations
          path: annotations
      name: istio-podinfo
    - name: istio-token
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            audience: istio-ca
            expirationSeconds: 43200
            path: istio-token
    - configMap:
        defaultMode: 420
        name: openshift-gw-ca-root-cert
      name: istiod-ca-cert
    - name: kube-api-access-fp5jr
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:31:28Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:31:26Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:31:51Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:31:51Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T21:31:26Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 128Mi
      containerID: cri-o://040d37622c67b40e8fc21225b4bc85e7d13e802e08035e8cb4eaec1f5c1c4ab4
      image: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:7d15cebf9b62f3f235c0eab5158ac8ff2fda86a1d193490dc94c301402c99da8
      imageID: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:0ae6667459cd9e9ea0549e27bb014c722924454bf594b31646df8df58be91cc5
      lastState:
        terminated:
          containerID: cri-o://4f33f43bf7a1c50599ace4b0a6d8ce58e760f576406b9bf0e92d532829415c18
          exitCode: 0
          finishedAt: "2026-07-02T21:31:44Z"
          reason: Completed
          startedAt: "2026-07-02T21:31:27Z"
      name: istio-proxy
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: 100m
          memory: 128Mi
      restartCount: 1
      started: true
      state:
        running:
          startedAt: "2026-07-02T21:31:44Z"
      user:
        linux:
          gid: 1000699999
          supplementalGroups:
          - 1000699999
          - 1000690000
          uid: 1000699999
      volumeMounts:
      - mountPath: /var/run/secrets/workload-spiffe-uds
        name: workload-socket
      - mountPath: /var/run/secrets/credential-uds
        name: credential-socket
      - mountPath: /var/run/secrets/workload-spiffe-credentials
        name: workload-certs
      - mountPath: /var/run/secrets/istio
        name: istiod-ca-cert
      - mountPath: /var/lib/istio/data
        name: istio-data
      - mountPath: /etc/istio/proxy
        name: istio-envoy
      - mountPath: /var/run/secrets/tokens
        name: istio-token
      - mountPath: /etc/istio/pod
        name: istio-podinfo
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-fp5jr
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.130.231
    hostIPs:
    - ip: 10.0.130.231
    observedGeneration: 1
    phase: Running
    podIP: 10.132.0.31
    podIPs:
    - ip: 10.132.0.31
    qosClass: Burstable
    startTime: "2026-07-02T21:31:26Z"
- apiVersion: v1
  kind: Pod
  metadata:
    annotations:
      istio.io/rev: openshift-gateway
      k8s.ovn.org/pod-networks: '{"default":{"ip_addresses":["10.134.0.38/23"],"mac_address":"0a:58:0a:86:00:26","gateway_ips":["10.134.0.1"],"routes":[{"dest":"10.132.0.0/14","nextHop":"10.134.0.1"},{"dest":"172.31.0.0/16","nextHop":"10.134.0.1"},{"dest":"169.254.0.5/32","nextHop":"10.134.0.1"},{"dest":"100.64.0.0/16","nextHop":"10.134.0.1"}],"ip_address":"10.134.0.38/23","gateway_ip":"10.134.0.1","role":"primary"}}'
      k8s.v1.cni.cncf.io/network-status: |-
        [{
            "name": "ovn-kubernetes",
            "interface": "eth0",
            "ips": [
                "10.134.0.38"
            ],
            "mac": "0a:58:0a:86:00:26",
            "default": true,
            "dns": {}
        }]
      openshift.io/scc: restricted-v2
      prometheus.io/path: /stats/prometheus
      prometheus.io/port: "15020"
      prometheus.io/scrape: "true"
      seccomp.security.alpha.kubernetes.io/pod: runtime/default
      security.openshift.io/validated-scc-subject-type: user
    creationTimestamp: "2026-07-02T22:34:48Z"
    generateName: router-gateway-2-openshift-default-54c789bdc6-
    generation: 1
    labels:
      gateway.istio.io/managed: istio.io-gateway-controller
      gateway.networking.k8s.io/gateway-name: router-gateway-2
      pod-template-hash: 54c789bdc6
      service.istio.io/canonical-name: router-gateway-2-openshift-default
      service.istio.io/canonical-revision: latest
      sidecar.istio.io/inject: "false"
    managedFields:
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.ovn.org/pod-networks: {}
      manager: ip-10-0-139-243
      operation: Update
      subresource: status
      time: "2026-07-02T22:34:48Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            .: {}
            f:istio.io/rev: {}
            f:prometheus.io/path: {}
            f:prometheus.io/port: {}
            f:prometheus.io/scrape: {}
          f:generateName: {}
          f:labels:
            .: {}
            f:gateway.istio.io/managed: {}
            f:gateway.networking.k8s.io/gateway-name: {}
            f:pod-template-hash: {}
            f:service.istio.io/canonical-name: {}
            f:service.istio.io/canonical-revision: {}
            f:sidecar.istio.io/inject: {}
          f:ownerReferences:
            .: {}
            k:{"uid":"27ac0905-39d7-4eed-86b8-69bc050b7289"}: {}
        f:spec:
          f:containers:
            k:{"name":"istio-proxy"}:
              .: {}
              f:args: {}
              f:env:
                .: {}
                k:{"name":"CA_ADDR"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"GOMAXPROCS"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"GOMEMLIMIT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"HOST_IP"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"INSTANCE_IP"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"ISTIO_CPU_LIMIT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:resourceFieldRef: {}
                k:{"name":"ISTIO_META_APP_CONTAINERS"}:
                  .: {}
                  f:name: {}
                k:{"name":"ISTIO_META_CLUSTER_ID"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_INTERCEPTION_MODE"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_MESH_ID"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_NODE_NAME"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"ISTIO_META_OWNER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_POD_PORTS"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"ISTIO_META_WORKLOAD_NAME"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"PILOT_CERT_PROVIDER"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"POD_NAME"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"POD_NAMESPACE"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"PROXY_CONFIG"}:
                  .: {}
                  f:name: {}
                  f:value: {}
                k:{"name":"SERVICE_ACCOUNT"}:
                  .: {}
                  f:name: {}
                  f:valueFrom:
                    .: {}
                    f:fieldRef: {}
                k:{"name":"TRUST_DOMAIN"}:
                  .: {}
                  f:name: {}
                  f:value: {}
              f:image: {}
              f:imagePullPolicy: {}
              f:name: {}
              f:ports:
                .: {}
                k:{"containerPort":15020,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":15021,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
                k:{"containerPort":15090,"protocol":"TCP"}:
                  .: {}
                  f:containerPort: {}
                  f:name: {}
                  f:protocol: {}
              f:readinessProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:resources:
                .: {}
                f:limits:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
                f:requests:
                  .: {}
                  f:cpu: {}
                  f:memory: {}
              f:securityContext:
                .: {}
                f:allowPrivilegeEscalation: {}
                f:capabilities:
                  .: {}
                  f:drop: {}
                f:privileged: {}
                f:readOnlyRootFilesystem: {}
                f:runAsGroup: {}
                f:runAsNonRoot: {}
                f:runAsUser: {}
              f:startupProbe:
                .: {}
                f:failureThreshold: {}
                f:httpGet:
                  .: {}
                  f:path: {}
                  f:port: {}
                  f:scheme: {}
                f:initialDelaySeconds: {}
                f:periodSeconds: {}
                f:successThreshold: {}
                f:timeoutSeconds: {}
              f:terminationMessagePath: {}
              f:terminationMessagePolicy: {}
              f:volumeMounts:
                .: {}
                k:{"mountPath":"/etc/istio/pod"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/etc/istio/proxy"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/lib/istio/data"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/credential-uds"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/istio"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/tokens"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/workload-spiffe-credentials"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
                k:{"mountPath":"/var/run/secrets/workload-spiffe-uds"}:
                  .: {}
                  f:mountPath: {}
                  f:name: {}
          f:dnsPolicy: {}
          f:enableServiceLinks: {}
          f:restartPolicy: {}
          f:schedulerName: {}
          f:securityContext:
            .: {}
            f:sysctls: {}
          f:serviceAccount: {}
          f:serviceAccountName: {}
          f:terminationGracePeriodSeconds: {}
          f:volumes:
            .: {}
            k:{"name":"credential-socket"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"istio-data"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"istio-envoy"}:
              .: {}
              f:emptyDir:
                .: {}
                f:medium: {}
              f:name: {}
            k:{"name":"istio-podinfo"}:
              .: {}
              f:downwardAPI:
                .: {}
                f:defaultMode: {}
                f:items: {}
              f:name: {}
            k:{"name":"istio-token"}:
              .: {}
              f:name: {}
              f:projected:
                .: {}
                f:defaultMode: {}
                f:sources: {}
            k:{"name":"istiod-ca-cert"}:
              .: {}
              f:configMap:
                .: {}
                f:defaultMode: {}
                f:name: {}
              f:name: {}
            k:{"name":"workload-certs"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
            k:{"name":"workload-socket"}:
              .: {}
              f:emptyDir: {}
              f:name: {}
      manager: kube-controller-manager
      operation: Update
      time: "2026-07-02T22:34:48Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:metadata:
          f:annotations:
            f:k8s.v1.cni.cncf.io/network-status: {}
      manager: multus-daemon
      operation: Update
      subresource: status
      time: "2026-07-02T22:34:48Z"
    - apiVersion: v1
      fieldsType: FieldsV1
      fieldsV1:
        f:status:
          f:conditions:
            k:{"type":"ContainersReady"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"Initialized"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodReadyToStartContainers"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
            k:{"type":"PodScheduled"}:
              f:observedGeneration: {}
            k:{"type":"Ready"}:
              .: {}
              f:lastProbeTime: {}
              f:lastTransitionTime: {}
              f:observedGeneration: {}
              f:status: {}
              f:type: {}
          f:containerStatuses: {}
          f:hostIP: {}
          f:hostIPs: {}
          f:observedGeneration: {}
          f:phase: {}
          f:podIP: {}
          f:podIPs:
            .: {}
            k:{"ip":"10.134.0.38"}:
              .: {}
              f:ip: {}
          f:startTime: {}
      manager: kubelet
      operation: Update
      subresource: status
      time: "2026-07-02T22:35:10Z"
    name: router-gateway-2-openshift-default-54c789bdc6-dhlsx
    namespace: kserve-ci-e2e-test
    ownerReferences:
    - apiVersion: apps/v1
      blockOwnerDeletion: true
      controller: true
      kind: ReplicaSet
      name: router-gateway-2-openshift-default-54c789bdc6
      uid: 27ac0905-39d7-4eed-86b8-69bc050b7289
    resourceVersion: "80274"
    uid: 2aac023f-0e70-4140-a697-e14f5f313a51
  spec:
    containers:
    - args:
      - proxy
      - router
      - --domain
      - $(POD_NAMESPACE).svc.cluster.local
      - --proxyLogLevel
      - warning
      - --proxyComponentLogLevel
      - misc:error
      - --log_output_level
      - default:info
      env:
      - name: PILOT_CERT_PROVIDER
        value: istiod
      - name: CA_ADDR
        value: istiod-openshift-gateway.openshift-ingress.svc:15012
      - name: POD_NAME
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.name
      - name: POD_NAMESPACE
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: metadata.namespace
      - name: INSTANCE_IP
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: status.podIP
      - name: SERVICE_ACCOUNT
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: spec.serviceAccountName
      - name: HOST_IP
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: status.hostIP
      - name: ISTIO_CPU_LIMIT
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.cpu
      - name: PROXY_CONFIG
        value: |
          {"discoveryAddress":"istiod-openshift-gateway.openshift-ingress.svc:15012","proxyHeaders":{"server":{"disabled":true},"envoyDebugHeaders":{"disabled":true},"metadataExchangeHeaders":{"mode":"IN_MESH"}}}
      - name: ISTIO_META_POD_PORTS
        value: '[]'
      - name: ISTIO_META_APP_CONTAINERS
      - name: GOMEMLIMIT
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.memory
      - name: GOMAXPROCS
        valueFrom:
          resourceFieldRef:
            divisor: "0"
            resource: limits.cpu
      - name: ISTIO_META_CLUSTER_ID
        value: Kubernetes
      - name: ISTIO_META_NODE_NAME
        valueFrom:
          fieldRef:
            apiVersion: v1
            fieldPath: spec.nodeName
      - name: ISTIO_META_INTERCEPTION_MODE
        value: REDIRECT
      - name: ISTIO_META_WORKLOAD_NAME
        value: router-gateway-2-openshift-default
      - name: ISTIO_META_OWNER
        value: kubernetes://apis/apps/v1/namespaces/kserve-ci-e2e-test/deployments/router-gateway-2-openshift-default
      - name: ISTIO_META_MESH_ID
        value: cluster.local
      - name: TRUST_DOMAIN
        value: cluster.local
      image: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:7d15cebf9b62f3f235c0eab5158ac8ff2fda86a1d193490dc94c301402c99da8
      imagePullPolicy: IfNotPresent
      name: istio-proxy
      ports:
      - containerPort: 15020
        name: metrics
        protocol: TCP
      - containerPort: 15021
        name: status-port
        protocol: TCP
      - containerPort: 15090
        name: http-envoy-prom
        protocol: TCP
      readinessProbe:
        failureThreshold: 4
        httpGet:
          path: /healthz/ready
          port: 15021
          scheme: HTTP
        periodSeconds: 15
        successThreshold: 1
        timeoutSeconds: 1
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: 100m
          memory: 128Mi
      securityContext:
        allowPrivilegeEscalation: false
        capabilities:
          drop:
          - ALL
        privileged: false
        readOnlyRootFilesystem: true
        runAsGroup: 1000699999
        runAsNonRoot: true
        runAsUser: 1000699999
      startupProbe:
        failureThreshold: 30
        httpGet:
          path: /healthz/ready
          port: 15021
          scheme: HTTP
        initialDelaySeconds: 1
        periodSeconds: 1
        successThreshold: 1
        timeoutSeconds: 1
      terminationMessagePath: /dev/termination-log
      terminationMessagePolicy: File
      volumeMounts:
      - mountPath: /var/run/secrets/workload-spiffe-uds
        name: workload-socket
      - mountPath: /var/run/secrets/credential-uds
        name: credential-socket
      - mountPath: /var/run/secrets/workload-spiffe-credentials
        name: workload-certs
      - mountPath: /var/run/secrets/istio
        name: istiod-ca-cert
      - mountPath: /var/lib/istio/data
        name: istio-data
      - mountPath: /etc/istio/proxy
        name: istio-envoy
      - mountPath: /var/run/secrets/tokens
        name: istio-token
      - mountPath: /etc/istio/pod
        name: istio-podinfo
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-crd9f
        readOnly: true
    dnsPolicy: ClusterFirst
    enableServiceLinks: true
    imagePullSecrets:
    - name: router-gateway-2-openshift-default-dockercfg-sf5t2
    nodeName: ip-10-0-139-243.ec2.internal
    preemptionPolicy: PreemptLowerPriority
    priority: 0
    restartPolicy: Always
    schedulerName: default-scheduler
    securityContext:
      fsGroup: 1000690000
      seLinuxOptions:
        level: s0:c26,c20
      seccompProfile:
        type: RuntimeDefault
      sysctls:
      - name: net.ipv4.ip_unprivileged_port_start
        value: "0"
    serviceAccount: router-gateway-2-openshift-default
    serviceAccountName: router-gateway-2-openshift-default
    terminationGracePeriodSeconds: 30
    tolerations:
    - effect: NoExecute
      key: node.kubernetes.io/not-ready
      operator: Exists
      tolerationSeconds: 300
    - effect: NoExecute
      key: node.kubernetes.io/unreachable
      operator: Exists
      tolerationSeconds: 300
    - effect: NoSchedule
      key: node.kubernetes.io/memory-pressure
      operator: Exists
    volumes:
    - emptyDir: {}
      name: workload-socket
    - emptyDir: {}
      name: credential-socket
    - emptyDir: {}
      name: workload-certs
    - emptyDir:
        medium: Memory
      name: istio-envoy
    - emptyDir: {}
      name: istio-data
    - downwardAPI:
        defaultMode: 420
        items:
        - fieldRef:
            apiVersion: v1
            fieldPath: metadata.labels
          path: labels
        - fieldRef:
            apiVersion: v1
            fieldPath: metadata.annotations
          path: annotations
      name: istio-podinfo
    - name: istio-token
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            audience: istio-ca
            expirationSeconds: 43200
            path: istio-token
    - configMap:
        defaultMode: 420
        name: openshift-gw-ca-root-cert
      name: istiod-ca-cert
    - name: kube-api-access-crd9f
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
        - configMap:
            items:
            - key: service-ca.crt
              path: service-ca.crt
            name: openshift-service-ca.crt
  status:
    conditions:
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:34:52Z"
      observedGeneration: 1
      status: "True"
      type: PodReadyToStartContainers
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:34:48Z"
      observedGeneration: 1
      status: "True"
      type: Initialized
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:35:10Z"
      observedGeneration: 1
      status: "True"
      type: Ready
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:35:10Z"
      observedGeneration: 1
      status: "True"
      type: ContainersReady
    - lastProbeTime: null
      lastTransitionTime: "2026-07-02T22:34:48Z"
      observedGeneration: 1
      status: "True"
      type: PodScheduled
    containerStatuses:
    - allocatedResources:
        cpu: 100m
        memory: 128Mi
      containerID: cri-o://049cb9f3f41a9a9e7afb74045bbd691f421d7472b391ac3426d35c91c1c244c5
      image: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:7d15cebf9b62f3f235c0eab5158ac8ff2fda86a1d193490dc94c301402c99da8
      imageID: registry.redhat.io/openshift-service-mesh/istio-proxyv2-rhel9@sha256:0ae6667459cd9e9ea0549e27bb014c722924454bf594b31646df8df58be91cc5
      lastState:
        terminated:
          containerID: cri-o://386513cbdbb82c7a2dd770798f9f086a96a8a1a879953696f305102cc5a84ec4
          exitCode: 0
          finishedAt: "2026-07-02T22:35:03Z"
          reason: Completed
          startedAt: "2026-07-02T22:34:51Z"
      name: istio-proxy
      ready: true
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: 100m
          memory: 128Mi
      restartCount: 1
      started: true
      state:
        running:
          startedAt: "2026-07-02T22:35:03Z"
      user:
        linux:
          gid: 1000699999
          supplementalGroups:
          - 1000699999
          - 1000690000
          uid: 1000699999
      volumeMounts:
      - mountPath: /var/run/secrets/workload-spiffe-uds
        name: workload-socket
      - mountPath: /var/run/secrets/credential-uds
        name: credential-socket
      - mountPath: /var/run/secrets/workload-spiffe-credentials
        name: workload-certs
      - mountPath: /var/run/secrets/istio
        name: istiod-ca-cert
      - mountPath: /var/lib/istio/data
        name: istio-data
      - mountPath: /etc/istio/proxy
        name: istio-envoy
      - mountPath: /var/run/secrets/tokens
        name: istio-token
      - mountPath: /etc/istio/pod
        name: istio-podinfo
      - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
        name: kube-api-access-crd9f
        readOnly: true
        recursiveReadOnly: Disabled
    hostIP: 10.0.139.243
    hostIPs:
    - ip: 10.0.139.243
    observedGeneration: 1
    phase: Running
    podIP: 10.134.0.38
    podIPs:
    - ip: 10.134.0.38
    qosClass: Burstable
    startTime: "2026-07-02T22:34:48Z"
kind: PodList
metadata:
  resourceVersion: "108289"