Open aflansburg opened 1 year ago
@aflansburg thanks for reporting this.
Which chart version is that? Is the readiness probe enabled?
I remember a similar issue existed before the C8 chart supported the readiness probe. The readiness has been enabled by default since 8.2.0
Describe the use case:
Use case:
ingress-gce
and container-native load balancing on Google Kubernetes engineWhen using
ingress-gce
on Google Kubernetes Engine for the identity service, Google somewhat herds you to container native load balancing, by creating the load balancing resources and network endpoint groups (NEGs) and even health checks required.However, backends can never become healthy as they always look to send health checks to port
80
. Backends and health checks and NEGs all point to port 80. You can add annotations and a BackendConfig CRD, however, the backend for port 80 will always look to port 80 for readiness probes.Without a ✅ from the backend, the Load Balancer will only respond with 502s.
Referencing potentially related issues: https://github.com/camunda/camunda-platform-helm/issues/707 https://github.com/camunda/camunda-platform-helm/issues/442
Here is an example of annotations that were used. While additional NEG was created for
metrics
, and a health check defined in aBackendConfig
CRD, it did not work as port 80->8082 is still being probed for readiness by the LB backend.Sidenote: ^ the above BackendConfig CRD for the zeebe gateway DOES actually mitigate this issue for that component.
Describe the enhancement/feature:
Identity service on port 80 responds to health checks at some path.
Desired outcome and acceptance tests:
GKE container-native load balancing configurations can receive a
200
response from the port enabling functionality of the LB.