Retry Internal Server Error(the server is currently unable to handle the request)

[tamayika]

When I'm applying helm chart, sometimes K8s server returns Internal Server Error.

Error: Failed to render chart: exit status 1: Error: Unable to continue with install: could not get information about the resource RoleBinding "namespace-labeler" in namespace "ci-namespace": an error on the server ("Internal Server Error: \"/apis/rbac.authorization.k8s.io/v1/namespaces/ci-namespace/rolebindings/namespace-labeler\": the server is currently unable to handle the request") has prevented the request from succeeding (get rolebindings.rbac.authorization.k8s.io namespace-labeler)

Current retry logic is limited to several internal server errors.

helm/pkg/cli/roundtripper.go

Lines 66 to 73 in 14d0c13

	// Matches messages like "etcdserver: leader changed"
	if strings.HasSuffix(ke.Message, "etcdserver: leader changed") {
	return rt.roundTrip(req, retry-1, resp)
	}
	// Matches messages like "rpc error: code = Unknown desc = raft proposal dropped"
	if strings.HasSuffix(ke.Message, "raft proposal dropped") {
	return rt.roundTrip(req, retry-1, resp)
	}

Is it reasonable to add the server is currently unable to handle the request error to retryable errors?
If ok, I will make a PR.

[sabre1041]

Possibly related #12976

[tamayika]

Thanks for the PR information.
But PR adds retry for 502 and 503, not 500.

[gjenkins8]

It might be okay to retry for GETs and HEADs on 500 (but not other http verbs). Might, because I can't help thinking your api-server might be overloaded in this case. And retying may be unhelpful to that situation. But with the exception of the additional load, retries are generally helpful, and can/should be added for idempotent requests.