Fix #5401: Polaris crash-loops on local Kubernetes: the OTel GCP resou

helm/polaris/templates/deployment.yaml

@@ -80,7 +80,7 @@ spec:
           {{- end }}
           image: "{{ tpl .Values.image.repository . }}:{{ tpl .Values.image.tag . | default .Chart.Version }}"
           imagePullPolicy: {{ tpl .Values.image.pullPolicy . }}
-          {{ if or .Values.storage.secret.name .Values.persistence.relationalJdbc.secret.name .Values.persistence.nosql.secret.name .Values.oidc.client.secret.name .Values.extraEnv -}}
+          {{ if or .Values.storage.secret.name .Values.persistence.relationalJdbc.secret.name .Values.persistence.nosql.secret.name .Values.oidc.client.secret.name (not .Values.tracing.enabled) .Values.tracing.disabledResourceProviders .Values.extraEnv -}}
           env:
             {{- include "polaris.secretToEnv" (list .Values.storage.secret "awsAccessKeyId" "polaris.storage.aws.access-key") | indent 12 -}}
             {{- include "polaris.secretToEnv" (list .Values.storage.secret "awsSecretAccessKey" "polaris.storage.aws.secret-key") | indent 12 -}}
@@ -90,6 +90,10 @@ spec:
             {{- include "polaris.secretToEnv" (list .Values.persistence.relationalJdbc.secret "jdbcUrl" "quarkus.datasource.jdbc.url") | indent 12 -}}
             {{- include "polaris.secretToEnv" (list .Values.persistence.nosql.secret "connectionString" "quarkus.mongodb.connection-string") | indent 12 -}}
             {{- include "polaris.secretToEnv" (list .Values.oidc.client.secret "key" "quarkus.oidc.credentials.secret") | indent 12 -}}
+            {{- if or (not .Values.tracing.enabled) .Values.tracing.disabledResourceProviders }}
+            - name: OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS
+              value: {{ default "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider" .Values.tracing.disabledResourceProviders | quote }}
+            {{- end }}
             {{- if .Values.extraEnv -}}
             {{- tpl (toYaml .Values.extraEnv) . | nindent 12 -}}
             {{- end -}}

helm/polaris/tests/deployment_test.yaml

@@ -382,11 +382,36 @@ tests:
           value: Always
 
   # spec.template.spec.containers[0].env
-  - it: should not set container env by default
+  - it: should set OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS by default when tracing is disabled
     template: deployment.yaml
+    asserts:
+      - contains:
+          path: spec.template.spec.containers[0].env
+          content:
+            name: OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS
+            value: "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider"
+  - it: should not set OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS when tracing is enabled
+    template: deployment.yaml
+    set:
+      tracing:
+        enabled: true
+        endpoint: "http://otlp-collector:4317"
     asserts:
       - notExists:
           path: spec.template.spec.containers[0].env
+  - it: should set OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS when tracing is enabled and disabledResourceProviders is set
+    template: deployment.yaml
+    set:
+      tracing:
+        enabled: true
+        endpoint: "http://otlp-collector:4317"
+        disabledResourceProviders: "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider"
+    asserts:
+      - contains:
+          path: spec.template.spec.containers[0].env
+          content:
+            name: OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS
+            value: "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider"
   - it: should set container env
     template: deployment.yaml
     set:
@@ -1144,11 +1169,17 @@ tests:
             configMapRef:
               name: polaris-env-configmap
 
-  - it: should not set any environment variables by default
+  - it: should set OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS as the only env var by default
     template: deployment.yaml
     asserts:
-      - notExists:
+      - lengthEqual:
+          path: spec.template.spec.containers[0].env
+          count: 1
+      - contains:
           path: spec.template.spec.containers[0].env
+          content:
+            name: OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS
+            value: "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider"
 
   - it: should not set envFrom by default
     template: deployment.yaml

helm/polaris/values.yaml

@@ -717,6 +717,12 @@ tracing:
   attributes:
     # service.name: my-polaris
     {}
+  # -- Comma-separated OTel resource provider class names to disable.
+  # When tracing is disabled (the default), the GCP resource detector is disabled
+  # automatically to prevent a ~135s startup stall on non-GCP Kubernetes clusters.
+  # Set this explicitly to override when tracing is enabled but the cluster is non-GCP.
+  # @section -- Observability
+  disabledResourceProviders: ""
 
 metrics:
   # -- Specifies whether metrics for the polaris server should be enabled. See [Metrics](https://polaris.apache.org/in-dev/unreleased/telemetry/#metrics) for details.

site/content/in-dev/unreleased/telemetry.md

@@ -117,6 +117,34 @@ Finally, two additional span attributes are added to all request parent spans:
 - `polaris.realm`: The unique identifier of the realm. Always set (unless the request failed because
   of a realm resolution error).
 
+### Slow startup in non-GCP Kubernetes clusters
+
+Polaris carries the GCP resource detector (`GCPResourceProvider`) on its classpath via a transitive
+dependency on GCS support. OpenTelemetry's autoconfigure runs resource providers during Quarkus
+startup even when tracing is disabled, so if `metadata.google.internal` resolves but silently drops
+packets (common in kind, k3d, minikube, and OrbStack Kubernetes), the `main` thread blocks for
+~135s waiting for a TCP connect timeout. Under the Helm chart's default liveness probe the pod
+crash-loops before startup completes.
+
+**Symptom:** no `started in` log line for ~135s, with no error or stack trace. A `SIGQUIT` thread
+dump shows `GCPMetadataConfig.fetchAttribute` on the `main` thread.
+
+**Helm chart:** when `tracing.enabled: false` (the default), the chart automatically sets
+`OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS` to skip the GCP detector. No action is needed.
+
+If you enable tracing but run on a non-GCP cluster, add the following to your Helm values:
+
+```yaml
+tracing:
+  disabledResourceProviders: "io.opentelemetry.contrib.gcp.resource.GCPResourceProvider"
+```
+
+Or set the environment variable directly:
+
+```shell
+OTEL_JAVA_DISABLED_RESOURCE_PROVIDERS=io.opentelemetry.contrib.gcp.resource.GCPResourceProvider
+```
+
 ### Troubleshooting Traces
 
 If the server is unable to publish traces, check first for a log warning message like the following: