client: call .Endpoints() in dial() in client/v3/client.go instead of accessing cfg.Endpoints directly

[yishuT]

etcd/client/v3/client.go

Line 299 in 0cdd558

initialEndpoints := strings.Join(c.cfg.Endpoints, ";")

accesses
endpoints without acquiring lock. Fix it to call Endpoints()

Fix #13201

[serathius]

I agree that this variable needs a lock to prevent data race with SetEndpoints, but I would also not expect that SetEndpoints should modify Config.
I would not expect config structs like client.v3.Config to be ever mutated by consumers. I expect that SetEndpoints was added later. Mutex withing Client doesn't directly specify which variables are protected by it.

Instead of of just adding lock, maybe we should consider adding dedicated endpoint field to Client struct and specifying explicitly that is protected by mutex?

[yishuT]

yeah, I agree. I only fixed what our race detector complained but didn't read the code in detail. There is one thing that if SetEndpoints is called for a new list of endpoints after client is init, now this client has two different list of endpoints, one in client.cfg, the other in client.endpoint.

[yishuT]

nvm, we don't do deep copy in SetEndpoints

[serathius]

LG

[yishuT]

any updates, or should I do something?

[spzala]

lgtm. Can you please squash commits into one? Thanks @yishuT

[yishuT]

squash commits

[spzala]

squash commits

Thanks!
CI is good, also tested locally.

[twz123]

Now that this PR is available in etcd 3.6, there's some GRPC warnings being issued when creating a client via client.New:

2025/10/15 08:16:40 WARNING: [core] [Channel #1 SubChannel #2] grpc: addrConn.createTransport failed to connect to {Addr: "127.0.0.1:2379", ServerName: "127.0.0.1", }. Err: connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:2379: operation was canceled"

Is that expected? Everything else works as intended, New returns without error, and the returned client is usable. However, a "warning" implies that something is not as it should be... 🤔

[twz123]

This calls client.resolver.SetEndpoints(eps), which hasn't been called before. I cannot really judge if this is a problem or not, but I observe GRPC warnings being logged with this:

2025/10/15 12:55:29 WARNING: [core] [Channel #1 SubChannel #2] grpc: addrConn.createTransport failed to connect to {Addr: "127.0.0.1:2379", ServerName: "127.0.0.1:2379", }. Err: connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:2379: operation was canceled"

or

2025/10/15 12:56:24 WARNING: [core] [Channel #1 SubChannel #2] grpc: addrConn.createTransport failed to connect to {Addr: "127.0.0.1:2379", ServerName: "127.0.0.1:2379", }. Err: connection error: desc = "transport: authentication handshake failed: context canceled"

Replacing this with the following makes them disappear again:

Suggested change

	client.SetEndpoints(cfg.Endpoints...)
	client.epMu.Lock()
	client.endpoints = cfg.Endpoints
	client.epMu.Unlock()