Skip to content

install-ate.sh: kind component deploys use GKE OTel endpoint #702

Closed

Description

Problem

On kind, --deploy-atenet and --deploy-ate-apiserver apply the base manifests instead of the kind Kustomize overlay. The base manifests set:

OTEL_EXPORTER_OTLP_ENDPOINT=http://opentelemetry-collector.gke-managed-otel.svc.cluster.local:4317

That collector exists on GKE, not in the kind stack, so the redeployed component cannot export OTLP telemetry. --deploy-atelet is not affected by this endpoint bug because it already renders a component-specific kind overlay.

Reproduction

hack/install-ate-kind.sh --deploy-ate-system
hack/install-ate-kind.sh --deploy-atenet
kubectl get deployment/atenet-router -n ate-system \
  -o jsonpath='{.spec.template.spec.containers[?(@.name=="atenet-router")].env[?(@.name=="OTEL_EXPORTER_OTLP_ENDPOINT")].value}'

The command returns the gke-managed-otel endpoint. The kind deployment should use:

http://opentelemetry-collector.otel-system.svc:4317

--deploy-ate-apiserver has the same problem.

Fix

Add component-specific kind overlays for atenet and ate-api-server, and use them when ATE_INSTALL_KIND=true. Keep the existing base-manifest path for GKE.

Acceptance criteria

  • Both partial kind deploys render the otel-system endpoint.
  • The ate-api-server partial overlay retains the kind metric interval and timeout settings.
  • Non-kind deploy behavior is unchanged.
  • The metrics E2E observes router telemetry after the SDK export interval.

Activity

  1. changed the title [-]e2e metrics: atenet-router metrics never reach the collector on kind[/-] [+]install-ate.sh: per-component deploys skip the kind overlay, silently breaking OTel export[/+] on Aug 2, 2026
  2. changed the title [-]install-ate.sh: per-component deploys skip the kind overlay, silently breaking OTel export[/-] [+]install-ate.sh: kind component deploys use GKE OTel endpoint[/+] on Aug 2, 2026
  3. added a commit that references this issue on Aug 4, 2026
    beacdeb
  4. orangeCatDeveloper commented on Aug 6, 2026

    @orangeCatDeveloper
    ContributorAuthor

    Fixed by #746: centralizing the OTLP endpoint in an ate-otel-config ConfigMap removes it from the workload manifests, so a kind component deploy can no longer revert it to the GKE collector.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    kind/bugSomething isn't working / bugfixes

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions