--- apiVersion: apps/v1 items: - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:14:55Z" 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: 68f5d8959c managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"0b2c84e7-86b7-43e7-b596-ad815acebd6b"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:14:55Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:16:55Z" name: auth-disabled-test-kserve-68f5d8959c namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: auth-disabled-test-kserve uid: 0b2c84e7-86b7-43e7-b596-ad815acebd6b resourceVersion: "30564" uid: 594d98e8-b847-491f-bca7-d49afb026bb0 spec: replicas: 1 selector: matchLabels: 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: 68f5d8959c template: metadata: 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: 68f5d8959c 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:14:55Z" 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: 5898f785fb managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"2d3259ac-03bd-42ae-b0d8-f179406fe26d"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:14:55Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:15:30Z" name: auth-disabled-test-kserve-router-scheduler-5898f785fb namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: auth-disabled-test-kserve-router-scheduler uid: 2d3259ac-03bd-42ae-b0d8-f179406fe26d resourceVersion: "29134" uid: 0067b8f6-2d4d-4411-8c67-36525950f914 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: auth-disabled-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 5898f785fb template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: auth-disabled-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 5898f785fb 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: auth-disabled-test-epp-sa serviceAccountName: auth-disabled-test-epp-sa terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:38:21Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: b6f9dd5f7 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"25fdd25b-bdc0-494d-9fd9-091157de6d8c"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:38:21Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:40:32Z" name: custom-route-timeout-test-kserve-b6f9dd5f7 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: custom-route-timeout-test-kserve uid: 25fdd25b-bdc0-494d-9fd9-091157de6d8c resourceVersion: "53250" uid: 4ace3925-32b3-4a6b-851c-7e654e33fa07 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: b6f9dd5f7 template: metadata: labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: b6f9dd5f7 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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: custom-route-timeout-test-kserve-self-signed-certs - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:38:22Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 65985dc45b managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"8ed9a5f3-b302-4058-9e59-c769dedb86cd"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:38:22Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:38:55Z" name: custom-route-timeout-test-kserve-router-scheduler-65985dc45b namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: custom-route-timeout-test-kserve-router-scheduler uid: 8ed9a5f3-b302-4058-9e59-c769dedb86cd resourceVersion: "51954" uid: 856ae2cd-e8e4-4c39-bfd0-ef380c3d8d47 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 65985dc45b template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: custom-route-timeout-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 65985dc45b 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 - custom-route-timeout-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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: custom-route-timeout-test-epp-sa serviceAccountName: custom-route-timeout-test-epp-sa terminationGracePeriodSeconds: 60 volumes: - name: tls-certs secret: defaultMode: 420 secretName: custom-route-timeout-test-kserve-self-signed-certs - emptyDir: {} name: tokenizer-uds - emptyDir: {} name: tokenizer-tmp - emptyDir: {} name: tokenizer-cache - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:21:01Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 5c57b59c67 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"7423aef6-87ce-4e5d-a3be-cb281cd37a44"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:21:01Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:23:02Z" name: llmisvc-model-fb-opt-125m-route-87882a8e-kserve-5c57b59c67 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc-model-fb-opt-125m-route-87882a8e-kserve uid: 7423aef6-87ce-4e5d-a3be-cb281cd37a44 resourceVersion: "36769" uid: 3d7e956d-0178-41fc-bb8a-a23beb61cfad spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 5c57b59c67 template: metadata: labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 5c57b59c67 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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: llmisva690bbc929faec8bc98c767f16c003c1-kserve-self-signed-certs - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:16:26Z" 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: 6694784b54 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"98638d0f-9512-4509-98e5-14865c1befce"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:16:26Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:17:56Z" name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-6694784b54 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve uid: 98638d0f-9512-4509-98e5-14865c1befce resourceVersion: "31661" uid: 58082a3a-5863-45c7-b4e7-9da5e3412076 spec: replicas: 1 selector: matchLabels: 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: 6694784b54 template: metadata: 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: 6694784b54 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt name: kserve-provision-location - mountPath: /etc/ssl/custom-certs name: cabundle-cert readOnly: true restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:16:26Z" 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: 6588f4987 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"07b4f646-3653-426e-9882-1ec59f2580e2"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:16:26Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:17:00Z" name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-router-scheduler-6588f4987 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc-model-fb-opt-125m-with-7ca60146-kserve-router-scheduler uid: 07b4f646-3653-426e-9882-1ec59f2580e2 resourceVersion: "30648" uid: 91372b18-811f-4ca0-9afc-bbd8326478d0 spec: replicas: 1 selector: matchLabels: 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: 6588f4987 template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" 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: 6588f4987 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: llmisvc-model-fb-opt-125m-with-7ca60146-epp-sa serviceAccountName: llmisvc-model-fb-opt-125m-with-7ca60146-epp-sa terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:33:12Z" 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: 55585bf4d9 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"208cf454-2b70-42ba-9bb3-2c453664c1f8"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:33:12Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:34:32Z" name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-55585bf4d9 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve uid: 208cf454-2b70-42ba-9bb3-2c453664c1f8 resourceVersion: "47882" uid: 820401f2-a724-4a90-9f00-21df066ddbb6 spec: replicas: 1 selector: matchLabels: 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: 55585bf4d9 template: metadata: 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: 55585bf4d9 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt name: kserve-provision-location - mountPath: /etc/ssl/custom-certs name: cabundle-cert readOnly: true restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:33:12Z" 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: 5d6c88c6f4 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"7bb204bb-a6c2-4eac-a0f7-f85970789783"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:33:12Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:33:47Z" name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-router-scheduler-5d6c88c6f4 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc-model-fb-opt-125m-with-ba4d693a-kserve-router-scheduler uid: 7bb204bb-a6c2-4eac-a0f7-f85970789783 resourceVersion: "47219" uid: 63184586-2b54-4322-b829-3b08a81be636 spec: replicas: 1 selector: matchLabels: 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: 5d6c88c6f4 template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" 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: 5d6c88c6f4 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: llmisvc-model-fb-opt-125m-with-ba4d693a-epp-sa serviceAccountName: llmisvc-model-fb-opt-125m-with-ba4d693a-epp-sa terminationGracePeriodSeconds: 60 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:54:26Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-pvc-router-manage-2577e794 app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 7dbcb75dbc managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"c97d0f34-4565-43ed-b20b-7e28ac516716"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:54:26Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:54:58Z" name: llmisvc1dc4269d1ada5f2d28562215d180c57f-kserve-router-scheduler-7dbcb75dbc namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvc1dc4269d1ada5f2d28562215d180c57f-kserve-router-scheduler uid: c97d0f34-4565-43ed-b20b-7e28ac516716 resourceVersion: "65131" uid: eb1a45f1-18a2-457e-8636-190d32b62411 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-pvc-router-manage-2577e794 app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 7dbcb75dbc template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-pvc-router-manage-2577e794 app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 7dbcb75dbc 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 - llmisvc-model-pvc-router-manage-2577e794-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 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 dnsPolicy: ClusterFirst restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: llmisvc-model-pvc-router-manage-2577e794-epp-sa serviceAccountName: llmisvc-model-pvc-router-manage-2577e794-epp-sa terminationGracePeriodSeconds: 60 volumes: - name: tls-certs secret: defaultMode: 420 secretName: llmisv1dc4269d1ada5f2d28562215d180c57f-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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:21:02Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 599658ff65 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"43b917fe-7e17-488b-831f-fafec25f64b0"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:21:02Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:21:35Z" name: llmisvca690bbc929faec8bc98c767f16c003c1-kserve-router-scheduler-599658ff65 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: llmisvca690bbc929faec8bc98c767f16c003c1-kserve-router-scheduler uid: 43b917fe-7e17-488b-831f-fafec25f64b0 resourceVersion: "35401" uid: 112023ff-5d0f-4641-9e6f-569aa6a74f4c spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 599658ff65 template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: llmisvc-model-fb-opt-125m-route-87882a8e app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 599658ff65 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 - llmisvc-model-fb-opt-125m-route-87882a8e-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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: llmisvc-model-fb-opt-125m-route-87882a8e-epp-sa serviceAccountName: llmisvc-model-fb-opt-125m-route-87882a8e-epp-sa terminationGracePeriodSeconds: 60 volumes: - name: tls-certs secret: defaultMode: 420 secretName: llmisva690bbc929faec8bc98c767f16c003c1-kserve-self-signed-certs - emptyDir: {} name: tokenizer-uds - emptyDir: {} name: tokenizer-tmp - emptyDir: {} name: tokenizer-cache - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:07:58Z" 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: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"3a62b7e4-d965-4759-a221-966bfd1ea2ae"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:istio.io/rev: {} f:prometheus.io/path: {} f:prometheus.io/port: {} f:prometheus.io/scrape: {} 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: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: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: "2026-07-08T07:07:58Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:19:00Z" name: router-gateway-1-openshift-default-799f46c59b namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-gateway-1-openshift-default uid: 3a62b7e4-d965-4759-a221-966bfd1ea2ae resourceVersion: "32812" uid: a6431df8-249b-4486-a74a-850f8322c05c spec: replicas: 1 selector: matchLabels: gateway.networking.k8s.io/gateway-name: router-gateway-1 pod-template-hash: 799f46c59b template: metadata: annotations: istio.io/rev: openshift-gateway prometheus.io/path: /stats/prometheus prometheus.io/port: "15020" prometheus.io/scrape: "true" 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" 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 dnsPolicy: ClusterFirst restartPolicy: Always schedulerName: default-scheduler securityContext: sysctls: - name: net.ipv4.ip_unprivileged_port_start value: "0" serviceAccount: router-gateway-1-openshift-default serviceAccountName: router-gateway-1-openshift-default terminationGracePeriodSeconds: 30 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T08:18:27Z" 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: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"4554083c-2685-4065-81de-1f9e7c9bcc02"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:istio.io/rev: {} f:prometheus.io/path: {} f:prometheus.io/port: {} f:prometheus.io/scrape: {} 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: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: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: "2026-07-08T08:18:27Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T08:18:50Z" name: router-gateway-2-openshift-default-54c789bdc6 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-gateway-2-openshift-default uid: 4554083c-2685-4065-81de-1f9e7c9bcc02 resourceVersion: "85055" uid: 4a74aae6-460b-4bc4-bc4f-0a20c49d9aa8 spec: replicas: 1 selector: matchLabels: gateway.networking.k8s.io/gateway-name: router-gateway-2 pod-template-hash: 54c789bdc6 template: metadata: annotations: istio.io/rev: openshift-gateway prometheus.io/path: /stats/prometheus prometheus.io/port: "15020" prometheus.io/scrape: "true" 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" 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 dnsPolicy: ClusterFirst restartPolicy: Always schedulerName: default-scheduler securityContext: sysctls: - name: net.ipv4.ip_unprivileged_port_start value: "0" serviceAccount: router-gateway-2-openshift-default serviceAccountName: router-gateway-2-openshift-default terminationGracePeriodSeconds: 30 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 status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T08:19:02Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: decode pod-template-hash: 6978d748b8 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"85b0811c-f56e-488a-b66b-319fc75ceb1f"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T08:19:02Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T08:22:22Z" name: router-with-refs-pd-test-kserve-6978d748b8 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-with-refs-pd-test-kserve uid: 85b0811c-f56e-488a-b66b-319fc75ceb1f resourceVersion: "87980" uid: 66fb1529-51cc-4b1e-b155-875b7876558b spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: decode pod-template-hash: 6978d748b8 template: metadata: labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: decode pod-template-hash: 6978d748b8 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 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 dnsPolicy: ClusterFirst 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/router-with-refs-pd-test-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: router-with-refs-pd-test-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 terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /var/run/kserve/tls name: tls-certs 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: router-with-refs-pd-test-kserve serviceAccountName: router-with-refs-pd-test-kserve terminationGracePeriodSeconds: 60 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: router-with-refs-pd-test-kserve-self-signed-certs - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T08:19:02Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-workload-prefill app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: prefill pod-template-hash: 56559b5fdf managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"f8c8c676-24f3-4173-8fd6-561c632e6436"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T08:19:02Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T08:22:22Z" name: router-with-refs-pd-test-kserve-prefill-56559b5fdf namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-with-refs-pd-test-kserve-prefill uid: f8c8c676-24f3-4173-8fd6-561c632e6436 resourceVersion: "87986" uid: 4ea7d5dc-71b8-4bfd-8bc6-9c429ce26728 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-workload-prefill app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: prefill pod-template-hash: 56559b5fdf template: metadata: labels: app.kubernetes.io/component: llminferenceservice-workload-prefill app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: prefill pod-template-hash: 56559b5fdf 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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: router-with-refs-pd-test-kserve-self-signed-certs - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T08:19:03Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 6ddb8947b5 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"93d73a57-13c4-434a-a4a8-bc0dfb26fd0f"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T08:19:03Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T08:19:37Z" name: router-with-refs-pd-test-kserve-router-scheduler-6ddb8947b5 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-with-refs-pd-test-kserve-router-scheduler uid: 93d73a57-13c4-434a-a4a8-bc0dfb26fd0f resourceVersion: "85896" uid: b2a90872-a2e7-4f27-968f-5f43b2f0851c spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 6ddb8947b5 template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-pd-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 6ddb8947b5 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 - router-with-refs-pd-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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: router-with-refs-pd-test-epp-sa serviceAccountName: router-with-refs-pd-test-epp-sa terminationGracePeriodSeconds: 60 volumes: - name: tls-certs secret: defaultMode: 420 secretName: router-with-refs-pd-test-kserve-self-signed-certs - emptyDir: {} name: tokenizer-uds - emptyDir: {} name: tokenizer-tmp - emptyDir: {} name: tokenizer-cache - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "2" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:55:58Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 794b6f7745 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"2f6d89a0-4f3e-4781-b925-68fde725f45b"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: 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: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: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: "2026-07-08T07:55:58Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:58:09Z" name: router-with-refs-test-kserve-794b6f7745 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-with-refs-test-kserve uid: 2f6d89a0-4f3e-4781-b925-68fde725f45b resourceVersion: "67754" uid: 071e58f0-2946-4c4e-a449-a998abd15a39 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 794b6f7745 template: metadata: labels: app.kubernetes.io/component: llminferenceservice-workload app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice kserve.io/component: workload llm-d.ai/role: both pod-template-hash: 794b6f7745 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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} terminationGracePeriodSeconds: 60 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: router-with-refs-test-kserve-self-signed-certs - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 - apiVersion: apps/v1 kind: ReplicaSet metadata: annotations: deployment.kubernetes.io/desired-replicas: "1" deployment.kubernetes.io/max-replicas: "1" deployment.kubernetes.io/revision: "1" creationTimestamp: "2026-07-08T07:55:59Z" generation: 1 labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 74dd96b744 managedFields: - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:metadata: f:annotations: .: {} f:deployment.kubernetes.io/desired-replicas: {} f:deployment.kubernetes.io/max-replicas: {} f:deployment.kubernetes.io/revision: {} 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":"2beb63eb-bbd9-4f46-b749-e6acf2d4dd89"}: {} f:spec: f:replicas: {} f:selector: {} f:template: f:metadata: f:annotations: .: {} f:app.kubernetes.io/version: {} f:certificates.kserve.io/expiration-v2: {} f:labels: .: {} f:app.kubernetes.io/component: {} f:app.kubernetes.io/name: {} f:app.kubernetes.io/part-of: {} f:pod-template-hash: {} 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: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: "2026-07-08T07:55:59Z" - apiVersion: apps/v1 fieldsType: FieldsV1 fieldsV1: f:status: f:availableReplicas: {} f:fullyLabeledReplicas: {} f:observedGeneration: {} f:readyReplicas: {} f:replicas: {} manager: kube-controller-manager operation: Update subresource: status time: "2026-07-08T07:56:33Z" name: router-with-refs-test-kserve-router-scheduler-74dd96b744 namespace: kserve-ci-e2e-test ownerReferences: - apiVersion: apps/v1 blockOwnerDeletion: true controller: true kind: Deployment name: router-with-refs-test-kserve-router-scheduler uid: 2beb63eb-bbd9-4f46-b749-e6acf2d4dd89 resourceVersion: "66503" uid: 0bbf19de-e126-4d6c-8580-b83de6f3a968 spec: replicas: 1 selector: matchLabels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 74dd96b744 template: metadata: annotations: app.kubernetes.io/version: 0.9.0 certificates.kserve.io/expiration-v2: "true" labels: app.kubernetes.io/component: llminferenceservice-router-scheduler app.kubernetes.io/name: router-with-refs-test app.kubernetes.io/part-of: llminferenceservice pod-template-hash: 74dd96b744 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 - router-with-refs-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 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 dnsPolicy: ClusterFirst 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@sha256:956644def59861b6c4d288d394adc3ba824655b2a247e0388911f52fbf9bf773 imagePullPolicy: IfNotPresent name: storage-initializer resources: limits: cpu: "1" memory: 24Gi requests: cpu: 100m memory: 100Mi terminationMessagePath: /dev/termination-log terminationMessagePolicy: FallbackToLogsOnError volumeMounts: - mountPath: /mnt/models name: kserve-provision-location restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: router-with-refs-test-epp-sa serviceAccountName: router-with-refs-test-epp-sa terminationGracePeriodSeconds: 60 volumes: - name: tls-certs secret: defaultMode: 420 secretName: router-with-refs-test-kserve-self-signed-certs - emptyDir: {} name: tokenizer-uds - emptyDir: {} name: tokenizer-tmp - emptyDir: {} name: tokenizer-cache - emptyDir: {} name: kserve-provision-location status: availableReplicas: 1 fullyLabeledReplicas: 1 observedGeneration: 1 readyReplicas: 1 replicas: 1 kind: ReplicaSetList metadata: resourceVersion: "98434"