How to detect whether an etcd member is a learner?

[kwohlfahrt]

If I add an etcd node to a cluster as a learner, the /health, /livez and /readyz endpoints all respond with a 200 OK response.

Unfortunately, when I then try to do any gRPC operations against this member, I get an error that the RPC is not supported by learners:

{"level":"warn","ts":"2024-06-25T15:03:10.490565+0200","logger":"etcd-client","caller":"[email protected]/retry_interceptor.go:62","msg":"retrying of unary invoker failed","target":"etcd-endpoints://0x140000141e0/localhost:2379","attempt":30,"error":"rpc error: code = Unavailable desc = etcdserver: rpc not supported for learner"}

Is there any easy health-check that will tell me whether the node I am connecting to is a learner or a full member?

[jmhbnz]

Hi @kwohlfahrt - Thanks for your question. For this use case you could wire up a health check based on etcdctl which can give some parseable json for the learner status, for example:

 ➜ ./etcdctl endpoint status -w json                                                                                                                                                                                                                                                
[{"Endpoint":"127.0.0.1:2379","Status":{"header":{"cluster_id":14841639068965178418,"member_id":10276657743932975437,"revision":1,"raft_term":2},"version":"3.6.0-alpha.0","dbSize":20480,"leader":10276657743932975437,"raftIndex":4,"raftTerm":2,"raftAppliedIndex":4,"dbSizeInUse
":16384,"storageVersion":"3.6.0"}}]

[kwohlfahrt]

Thanks for the clarification - this is indeed more or less what I'm doing at the moment as a workaround.

If there is no HTTP endpoint, I'll probably open a feature request for this then, because it's stopping me from integrating with some third-party tools that expect HTTP endpoint for health/readiness checks. IMHO it doesn't make sense to call an etcd node "ready" if it can't respond to any requests.