Helm Chart

Deploy Spice.ai Enterprise on Kubernetes with the official Helm chart.

The Spice.ai Enterprise Helm chart deploys the Spice runtime as a Kubernetes Deployment or StatefulSet with a ConfigMap-mounted Spicepod configuration.

Install

The chart is published to the AWS Marketplace ECR registry for the enterprise and enterprise-models image variants, with the image pre-configured. The -sysalloc image has no chart of its own: set image.tag to run it. Subscribe to the Enterprise listing first (see AWS Marketplace), then authenticate Helm against the registry:

aws ecr get-login-password --region us-east-1 \
  | helm registry login --username AWS --password-stdin 709825985650.dkr.ecr.us-east-1.amazonaws.com

The login is valid for 12 hours; re-run it when the token expires.

helm install spiceai \
  oci://709825985650.dkr.ecr.us-east-1.amazonaws.com/spice-ai/spiceai-enterprise-byol \
  --version 2.4.1-enterprise-models-helm \
  --set spicepod.name=my-app

Values Reference

ParameterDescriptionDefault
image.repositoryContainer image repository709825985650.dkr.ecr.us-east-1.amazonaws.com/spice-ai/spiceai-enterprise-byol
image.tagContainer image tagThe chart variant's image tag, such as 2.4.1-enterprise-models
imagePullSecretsImage pull secrets for the podNone
replicaCountNumber of replicas1
strategyUpdate strategy for the DeploymentNone (Kubernetes default)
updateStrategyUpdate strategy for the StatefulSetNone (Kubernetes default)
commandContainer command that starts spicedspiced with HTTP on 8090, metrics on 9090, and Arrow Flight on 50051
livenessProbeLiveness probe for the runtime containerHTTP GET /health on port 8090
readinessProbeReadiness probe for the runtime containerHTTP GET /v1/ready on port 8090
startupProbeStartup probe for the runtime containerHTTP GET /health on port 8090
serviceAccount.createCreate a ServiceAccountfalse
serviceAccount.nameServiceAccount nameNone
serviceAccount.annotationsAnnotations for the ServiceAccount{}
service.typeKubernetes Service typeNone (Kubernetes default, ClusterIP)
service.additionalAnnotationsAnnotations for the Service{}
service.loadBalancerSourceRangesSource IP ranges allowed to reach a LoadBalancer ServiceNone
service.portsService ports, which replace the default list8090 (http), 9090 (metrics), 50051 (flight)
service.selectorService selector, which replaces the default selectorThe chart's labels, additionalLabels, and app: <release name>
stateful.enabledUse a StatefulSet with PVCfalse
stateful.storageClassStorageClass for StatefulSet PVCstandard
stateful.sizePVC size1Gi
stateful.mountPathMount path in container/data
monitoring.podMonitor.enabledCreate a PodMonitor for Prometheusfalse
monitoring.podMonitor.additionalLabelsLabels for the PodMonitor{}
additionalLabelsLabels added to all resources{}
nameOverrideName used in place of the chart name in resource names and in the app.kubernetes.io/name and app.kubernetes.io/part-of labelsThe chart name
fullnameOverrideName of the Deployment or StatefulSet and the Service, the ConfigMap prefix, and the default name of a created ServiceAccount, truncated to 63 characters<release name>-<name>, or the release name when it contains the name
namespaceOverrideNamespace for the workload, the Service, the ConfigMap, and the ServiceAccount. The PodMonitor stays in the release namespaceThe release namespace
selectorOverrideLabels that replace app.kubernetes.io/name and app.kubernetes.io/instance in the pod selector and in the labels of every resource. The chart adds app: <release name> to the selectorNone
additionalEnvExtra environment variables[]
resourcesCPU/memory requests and limits{}
nodeSelectorNode selector for the podNone
tolerationsTolerations for the podNone
affinityAffinity rules for the podNone
volumesAdditional volumes[]
volumeMountsAdditional volume mounts[]
spicepodSpicepod configuration (inlined into ConfigMap)name: app, version: v1, kind: Spicepod, and no datasets

Spicepod Configuration

The Spicepod spec is inlined directly in the Helm values and mounted as a ConfigMap:

spicepod:
  name: my-app
  version: v1
  kind: Spicepod
  datasets:
    - from: s3://my-bucket/data/
      name: my_data
      params:
        file_format: parquet
      acceleration:
        enabled: true
        engine: cayenne

StatefulSet with Persistent Storage

Enable a StatefulSet with a PersistentVolumeClaim for durable storage:

stateful:
  enabled: true
  storageClass: gp3
  size: 50Gi
  mountPath: /data

Update Strategy

Both keys are unset by default, so Kubernetes applies its own default strategy. Which one the chart reads depends on stateful.enabled.

With stateful.enabled: false the chart renders a Deployment and reads strategy:

strategy:
  type: Recreate

With stateful.enabled: true the chart renders a StatefulSet and reads updateStrategy:

updateStrategy:
  type: RollingUpdate

The key that does not match stateful.enabled is ignored, so setting strategy on a StatefulSet deployment has no effect. Each value is passed through to the workload spec unchanged, so any field the Kubernetes strategy type accepts (for example rollingUpdate.maxUnavailable or rollingUpdate.partition) can be set alongside type.

Health Probes

The chart sets a liveness, readiness, and startup probe on the runtime container. Each probe uses these defaults:

livenessProbe:
  httpGet:
    path: /health
    port: 8090
  timeoutSeconds: 1
  periodSeconds: 10
  failureThreshold: 3

The readiness probe calls /v1/ready instead of /health. Helm merges your values with these defaults, so you can override one field. For example, give a large dataset more time to load before the kubelet restarts the container:

startupProbe:
  failureThreshold: 60

When a probe sets exec, tcpSocket, or grpc, the chart removes the default httpGet handler from that probe.

Container Command

By default, the container runs this command:

command:
  - /usr/local/bin/spiced
  - --http
  - 0.0.0.0:8090
  - --metrics
  - 0.0.0.0:9090
  - --flight
  - 0.0.0.0:50051

Set command to replace the full list, for example to add a spiced flag. Include every flag you need, because the chart does not merge your list with the default. The container ports stay 8090 (HTTP), 9090 (metrics), and 50051 (Flight), and the default probes call port 8090. Keep each address on its port.

Service Exposure

The Service exposes ports 8090 (http), 9090 (metrics), and 50051 (flight). To expose the runtime through a cloud load balancer and allow only some source ranges:

service:
  type: LoadBalancer
  loadBalancerSourceRanges:
    - 10.0.0.0/8

Pod Scheduling and Image Pull Secrets

nodeSelector, tolerations, affinity, and imagePullSecrets are passed to the pod spec unchanged:

nodeSelector:
  node-pool: spice
tolerations:
  - key: dedicated
    operator: Equal
    value: spice
    effect: NoSchedule
imagePullSecrets:
  - name: my-registry-secret

Resource Limits

resources:
  requests:
    cpu: 200m
    memory: 1Gi
  limits:
    cpu: "2"
    memory: 4Gi

Prometheus Monitoring

monitoring:
  podMonitor:
    enabled: true
    additionalLabels:
      release: prometheus

ServiceAccount for Cloud IAM

AWS IRSA (EKS)

serviceAccount:
  create: true
  name: spiceai
  annotations:
    eks.amazonaws.com/role-arn: arn:aws:iam::123456789012:role/my-spice-role

See AWS IRSA for full setup instructions.

GKE Workload Identity

serviceAccount:
  create: true
  annotations:
    iam.gke.io/gcp-service-account: my-gcp-sa@my-project.iam.gserviceaccount.com

Azure Workload Identity (AKS)

serviceAccount:
  create: true
  annotations:
    azure.workload.identity/client-id: <AZURE_CLIENT_ID>