agentleFS
Aidas-devAidas-dev/k8s-agent-skills/skills/kserve-helm/SKILL.md

kserve-helm

Skill4 starsChanged 5 months agoMITView on GitHub

Deploy KServe on Kubernetes via Helm — 10 charts, Serverless/RawDeployment/ModelMesh modes, KServe/ModelMesh/annotator/router configurations.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Default mode. Auto-scaling to zero via Knative.

    line 174

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

KServe — Helm Deployment

Repo: https://kserve.github.io/helm-charts
Charts: 10 (see table below)
Latest: v0.18.0

Charts (10)

ChartNamespacePurpose
kserve-crdkserveCore CRDs only
kserve-resourceskserveController + webhook + storage init
kserve-llmisvc-crdkserveLLMInferenceService CRD only
kserve-llmisvc-resourceskserveLLMInferenceService controller
kserve-localmodel-crdkserveLocalModelCache + LocalModelNode CRDs
kserve-localmodel-resourceskserveLocalModel controller + node agent
kserve-runtime-configskserveDefault ServingRuntime configs
kserve-crd-minimalkserveMin CRDs (InferenceService only)
kserve-resources-minimalkserveMin controller (no Mesh/Knative)
kserve-runtime-configs-minimalkserveMinimal runtime configs

Prerequisites

DependencyVersionRequired For
Kubernetes≥ 1.26All
Cert-Manager≥ 1.12Controller webhook
Istio≥ 1.20Standard mode (with ingress gateway)
Knative Serving≥ 1.16Serverless mode (optional, auto-scaling to zero)
Gateway API≥ 1.2Gateway API ingress (alternative to Istio)

Quick Install

Standard Mode (Knative + Istio)

# 1. Install Knative (if needed)
kubectl apply -f https://github.com/knative/serving/releases/download/knative-v1.16.0/serving-crds.yaml
kubectl apply -f https://github.com/knative/serving/releases/download/knative-v1.16.0/serving-core.yaml
kubectl apply -f https://github.com/knative/net-istio/releases/download/knative-v1.16.0/net-istio.yaml

# 2. Install KServe
helm repo add kserve https://kserve.github.io/helm-charts
helm repo update

# CRDs
helm install kserve-crd oci://ghcr.io/kserve/charts/kserve-crd \
  --version v0.18.0 --namespace kserve --create-namespace

# Controllers
helm install kserve-resources oci://ghcr.io/kserve/charts/kserve-resources \
  --version v0.18.0 --namespace kserve

# Runtime configs
helm install kserve-runtime-configs oci://ghcr.io/kserve/charts/kserve-runtime-configs \
  --version v0.18.0 --namespace kserve

Raw Deployment Mode (No Mesh)

# Minimal CRDs (InferenceService only)
helm install kserve-crd oci://ghcr.io/kserve/charts/kserve-crd-minimal \
  --version v0.18.0 --namespace kserve --create-namespace

# Minimal controller
helm install kserve-resources oci://ghcr.io/kserve/charts/kserve-resources-minimal \
  --version v0.18.0 --namespace kserve

# Minimal runtime configs
helm install kserve-runtime-configs oci://ghcr.io/kserve/charts/kserve-runtime-configs-minimal \
  --version v0.18.0 --namespace kserve

LLM Inference Service (Optional)

helm install kserve-llmisvc-crd oci://ghcr.io/kserve/charts/kserve-llmisvc-crd \
  --version v0.18.0 --namespace kserve

helm install kserve-llmisvc-resources oci://ghcr.io/kserve/charts/kserve-llmisvc-resources \
  --version v0.18.0 --namespace kserve

LocalModel (Optional)

helm install kserve-localmodel-crd oci://ghcr.io/kserve/charts/kserve-localmodel-crd \
  --version v0.18.0 --namespace kserve

helm install kserve-localmodel-resources oci://ghcr.io/kserve/charts/kserve-localmodel-resources \
  --version v0.18.0 --namespace kserve

Key Values (kserve-resources)

ParameterDefaultDescription
kserve.controller.deploymentModeServerlessServerless, RawDeployment, ModelMesh
kserve.controller.gateway.ingressGateway.enabledtrueEnable ingress gateway config
kserve.controller.gateway.ingressGateway.gatewayServiceistio-ingressgateway.istio-systemIngress gateway service
kserve.controller.gateway.ingressGateway.localGatewayServiceknative-local-gateway.knative-servingLocal gateway for mesh
kserve.controller.gateway.clusterLocalGateway.enabledtrueEnable cluster-local gateway
kserve.controller.gateway.clusterLocalGateway.serviceknative-local-gateway.knative-servingLocal gateway service

Controller

ParameterDefaultDescription
kserve.controller.imagekserve/kserve-controllerController image
kserve.controller.tagv0.18.0Image tag
kserve.controller.replicas1Controller replicas
kserve.controller.resources{cpu: 100m/500m, mem: 256Mi/1Gi}Container resources
kserve.controller.nodeSelector{}Node selector
kserve.controller.tolerations[]Tolerations
kserve.controller.affinity{}Pod affinity
kserve.controller.topologySpreadConstraints[]Topology spread
kserve.controller.pdb.enabledfalsePDB
kserve.controller.pdb.maxUnavailable1PDB max unavailable
kserve.controller.env[]Extra env vars
kserve.controller.args[]Extra args

Webhook

ParameterDefaultDescription
kserve.webhook.imagekserve/kserve-controllerWebhook image
kserve.webhook.tagv0.18.0Image tag
kserve.webhook.replicas1Webhook replicas
kserve.webhook.resources{cpu: 100m/500m, mem: 256Mi/1Gi}Container resources
kserve.webhook.nodeSelector{}Node selector
kserve.webhook.tolerations[]Tolerations
kserve.webhook.affinity{}Pod affinity

Storage Initializer

ParameterDefaultDescription
kserve.storageInitializer.imagekserve/storage-initializerInit container image
kserve.storageInitializer.tagv0.18.0Image tag
kserve.storageInitializer.resources{cpu: 100m/1, mem: 200Mi/1Gi}Init container resources
kserve.storageInitializer.env[]Extra env vars

Agent

ParameterDefaultDescription
kserve.agent.imagekserve/agentAgent image
kserve.agent.tagv0.18.0Image tag

Router

ParameterDefaultDescription
kserve.router.imagekserve/routerRouter image
kserve.router.tagv0.18.0Image tag

Global

ParameterDefaultDescription
global.imagePullSecrets[]Image pull secrets
global.customMetrics.enabledfalseEnable custom metrics adapter

Deployment Modes

Standard Mode (Serverless - Knative)

Default mode. Auto-scaling to zero via Knative. Requires Knative Serving + Istio.

helm install kserve-resources oci://ghcr.io/kserve/charts/kserve-resources \
  --version v0.18.0 --namespace kserve \
  --set kserve.controller.deploymentMode=Serverless

Characteristics:

  • Auto-scaling to zero (cold start)
  • Traffic splitting (canary)
  • Revision-based model versioning
  • Requires Knative + Istio

Raw Deployment Mode

Direct Deployments without Knative. No auto-scaling to zero. Simpler stack.

helm install kserve-resources oci://ghcr.io/kserve/charts/kserve-resources \
  --version v0.18.0 --namespace kserve \
  --set kserve.controller.deploymentMode=RawDeployment

Characteristics:

  • No Knative dependency
  • Standard K8s Deployments + Services
  • HPA for autoscaling (minReplicas ≥ 1)
  • Lighter resource usage

ModelMesh Mode

Multi-model serving on shared runtimes. Requires ServingRuntime CRDs.

helm install kserve-resources oci://ghcr.io/kserve/charts/kserve-resources \
  --version v0.18.0 --namespace kserve \
  --set kserve.controller.deploymentMode=ModelMesh

Characteristics:

  • Multiple models per pod (multiModel)
  • ServingRuntime templates
  • Model pooling
  • Efficient GPU utilization

Production Values Example

Standard Mode Production

kserve:
  controller:
    deploymentMode: Serverless
    replicas: 3
    resources:
      requests:
        cpu: 500m
        memory: 1Gi
      limits:
        cpu: 2
        memory: 4Gi
    env:
      - name: ENABLE_GPU
        value: "true"
    affinity:
      podAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
                - key: app
                  operator: In
                  values:
                    - kserve-controller
            topologyKey: kubernetes.io/hostname
    pdb:
      enabled: true
      maxUnavailable: 1

  webhook:
    replicas: 3
    resources:
      requests:
        cpu: 200m
        memory: 512Mi
      limits:
        cpu: 500m
        memory: 1Gi

  storageInitializer:
    env:
      - name: AWS_ENDPOINT_URL
        value: s3.us-east-1.amazonaws.com
      - name: AWS_REGION
        value: us-east-1

Raw Deployment Mode

kserve:
  controller:
    deploymentMode: RawDeployment
    gateway:
      ingressGateway:
        enabled: false
      clusterLocalGateway:
        enabled: false

Upgrading

helm repo update

# Upgrade CRDs first
helm upgrade kserve-crd oci://ghcr.io/kserve/charts/kserve-crd \
  --version v0.18.0 --namespace kserve

# Upgrade controller
helm upgrade kserve-resources oci://ghcr.io/kserve/charts/kserve-resources \
  --version v0.18.0 --namespace kserve

# Upgrade runtime configs
helm upgrade kserve-runtime-configs oci://ghcr.io/kserve/charts/kserve-runtime-configs \
  --version v0.18.0 --namespace kserve

Uninstalling

# Uninstall resources (reverses operation order)
helm uninstall kserve-runtime-configs --namespace kserve
helm uninstall kserve-localmodel-resources --namespace kserve
helm uninstall kserve-localmodel-crd --namespace kserve
helm uninstall kserve-llmisvc-resources --namespace kserve
helm uninstall kserve-llmisvc-crd --namespace kserve
helm uninstall kserve-resources --namespace kserve

# Uninstall CRDs last
helm uninstall kserve-crd --namespace kserve

# Delete namespace
kubectl delete namespace kserve

⚠️ WARNING: Uninstalling kserve-crd deletes ALL CRDs and cascades to delete ALL InferenceServices across ALL namespaces.

Common Mistakes

  • CRDs not installed first — The controller chart fails without CRDs. Always install kserve-crd first.
  • Missing cert-manager — KServe webhook requires cert-manager ≥ 1.12. Without it, controller pods crash-loop with webhook TLS errors.
  • Deployment mode mismatch — Serverless mode requires Knative installed. RawDeployment does not. Set correctly before install.
  • Knative + Istio version mismatch — KServe v0.18.0 requires Knative ≥ 1.16 and Istio ≥ 1.20. Version mismatch causes networking errors.
  • Stale webhook certs — After upgrade, webhook certificates may need re-issuance. Restarting the webhook pod usually resolves.
  • Gateway service exists — If istio-ingressgateway.istio-system doesn't exist in Standard mode, InferenceServices get Failed to get gateway status.
  • Localmodel with non-NVMe storage — LocalModel cache works best with NVMe. Regular disks may cause model loading performance issues.
  • ModelMesh with single model — ModelMesh expects multiModel: true on ServingRuntime. For single-model, use Standard mode.
  • v0.19.0-rc0 caution — Release candidate. Test for production before using.
  • OCI chart vs Helm repo — Use oci://ghcr.io/kserve/charts/ for v0.18.0+ charts. Older versions used the HTTPS helm repo.
  • Annotations not propagated — Some predictor annotations (e.g., serving.kserve.io/deploymentMode) must be on the InferenceService, not the pod template.
  • Storage initializer timeout — Large models (>10GB) may exceed the default init container timeout. Increase via kserve.storageInitializer.env.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.