Uh oh!
There was an error while loading. Please reload this page.
[DO NOT MERGE] Add hyperpod-ray-endpoint-operator to HyperPod Helm chart - #445
Conversation
0aecf49 to
9b37bc8Comparepsnilesh
commented
Aug 10, 2026
In this case, which namespace is the endpoint operator installed to? |
psnilesh
commented
Aug 10, 2026
We also need to know the version that's being used from the logs. Should we add the helm version (i.e sagemaker-hyperpod-cli repo release version) as an env variable to all operator pods and log it at startup? |
jchatter321
commented
Aug 10, 2026
The operator will be installed in hyperpod-ray. |
jchatter321
commented
Aug 10, 2026
Discussed offline, will address this in another PR. |
9b37bc8 to
68a57f1Compare68a57f1 to
37286c0Compare37286c0 to
3f6acbcCompareUh oh!
There was an error while loading. Please reload this page.
Summary
Adds the hyperpod-ray-endpoint-operator as a sub-chart to the HyperPodHelmChart. The chart can be installed with the following command -
To install the chart in a specific namespace -
To install with KMS-based JWT signing (instead of K8s Secrets):
To install with a custom admin group:
To install with a custom image override (e.g., for testing with a specific digest):
To install from the parent chart:
HyperPod Ray Endpoint Operator — Installation Defaults
Required at install time
regionimage.overrideis set)domainspaces.example.com)Namespace
namespacehyperpod-rayImage
hyperpod-ray-endpoint-operator(hardcoded)1.0.173.0_1.0.18.0image.override""(off) — when set, bypasses all resolutionJWT signing
kmsKeyArn""(empty)Controller Manager
manager.enabledtruemanager.replicas1manager.image.pullPolicyIfNotPresentmanager.resources.limitsmanager.resources.requestsmanager.nodeSelectoramd64,sagemaker.amazonaws.com/compute-type: hyperpodmanager.terminationGracePeriodSeconds10--leader-elect)--enable-extension-api)Auth Middleware
authMiddleware.enabledtrueauthMiddleware.replicas2authMiddleware.image.pullPolicyIfNotPresentauthMiddleware.resources.limitsauthMiddleware.resources.requestsauthMiddleware.env.sessionTTL6hauthMiddleware.env.cookieSecuretrueauthMiddleware.env.cookieSameSitelaxauthMiddleware.nodeSelectoramd64,sagemaker.amazonaws.com/compute-type: hyperpodauthMiddleware.terminationGracePeriodSeconds30Access Control
adminGroupsystem:mastersMetrics & Monitoring
metrics.enabletruemetrics.port8443metrics.securetrue(HTTPS)prometheus.enablefalse(no ServiceMonitor)Security (all components)
runAsNonRoot: true, UID/GID65532readOnlyRootFilesystem: trueallowPrivilegeEscalation: falseRuntimeDefaultParent chart
hyperpod-ray-endpoint-operator.enabledfalse(must explicitly enable)Testing
Verified with
helm templateacross multiple regions and override scenarios. Deployed to local cluster against different namespace and ran internal e2e tests to validate functionality.