client.close() occasionally takes 10s

Hello,
Occasionally, but quite often, I have a case where `client.close()` takes 10 seconds (between 10.0 and 10.2 to be more specific).
I tried lowering the client timeout (assuming it may be connected), but it doesn't seem to have any influence.

I guess problem may be connected with gRPC client closing, but not sure how to troubleshoot it further.

ETCD logs don't show anything different from the case where `close()` happens immediately...

I would be grateful for advices...

Start at the client.close() entry point and trace the gRPC client shutdown path, comparing cases that close immediately with those that take 10.0–10.2 seconds. Check whether the configured client timeout affects this path and compare the corresponding etcd logs; done means the cause of the delay is identified and the close behavior is corrected or clearly documented.

Written by the indexing model from the issue text.