Open meimeitou opened 6 years ago
I got the same problem, seems you need to add yourself's reconnection code.
I have the same problem when using authentication. etcd's auth token expires after about 5 minutes, and after then, all API calls fail.
I've ended up writing an adaptor that reconnects & reauthenticates with specific grpc errors. https://github.com/lablup/backend.ai-common/commit/0526a3f7a2800966f5350cbac264561d0b657f8e
Note: When initializing the etcd3 client, the first authentication routine does not apply _handle_errors()
so I had to catch "UNAVAILABLE" exception instead of etcd3.exceptions.ConnectionFailedError
.
connected to server and few minutes later call client.get('/xxx'), got error: