Liveness/Readiness check for external Endpoints Services

[MartinEmrich]

What would you like to be added:

Liveness/Readiness checks for external Endpoints.

As I understand, kube-proxy does load-balancing over Endpoints with multiple targets like

apiVersion: v1
kind: Endpoints
metadata:
  name: someClusterOfAServiceOutsideOfTheKubernetesCluster
subsets:
- addresses:
  - ip: 10.32.123.456
  - ip: 10.32.234.567
  - ip: 10.32.345.678
  ports:
  - port: 1234

But the documentation is not clear (to me) what happens if one of these three backends does not work, but I assume that addresses with the port closed are excluded from load balancing? But what about retry intervals or check intervals?

Why would this be nice:

Adding liveness/readiness probes to Endpoints would make this configurable, e.g. to check either simple reachability (TCP) or periodically calling a liveness route (e.g. HTTP call to /myservice/alive an checking for 200 OK).

This might look like:

apiVersion: v1
kind: Endpoints
metadata:
  name: someClusterOfAServiceOutsideOfTheKubernetesCluster
subsets:
- addresses:
  - ip: 10.32.123.456
  - ip: 10.32.234.567
  - ip: 10.32.345.678
  ports:
  - port: 1234
  readinessProbe:
    httpGet:
      path: /myservice/alive
    periodSeconds: 15

Maybe this would be out of scope for Kubernetes (and best practice would be to use a loadbalancer like haproxy in a small pod), but maybe the Idea is not bad?

[MartinEmrich]

@k8s-ci-robot Here you go :)

/sig network

[athenabot]

/triage unresolved

Comment /remove-triage unresolved when the issue is assessed and confirmed.

🤖 I am a bot run by vllry. 👩‍🔬

[vllry]

Only ready pods have an endpoint added to the Endpoints object (there is no concept of health checks at the service/endpoint level). This allows consumers of endpoints to avoid the need to health check. So I'm afraid this would be out-of-scope feature request.

There's lots of ways you could accomplish this, your best bet would be a load balancer rather than external Endpoints. But there's many ways you could customize things like adding a service-equivalent controller with this functionality, etc. ;)

/close

[k8s-ci-robot]

@vllry: Closing this issue.

In response to this:

Only ready pods have an endpoint added to the Endpoints object (there is no concept of health checks at the service/endpoint level). This allows consumers of endpoints to avoid the need to health check. So I'm afraid this would be out-of-scope feature request.

There's lots of ways you could accomplish this, your best bet would be a load balancer rather than external Endpoints. But there's many ways you could customize things like adding a service-equivalent controller with this functionality, etc. ;)

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.