when the grpc server restart,the grpcClient is not available,I have to restart the grpc client

[champion1Ping]

status=FAILURE, cause=[io.grpc.StatusRuntimeException: UNAVAILABLE: upstream connect error or disconnect/reset before headers. retried and the latest reset reason: connection failure

[champion1Ping]

the server is implemented in go language

[fengli79]

In general, when the gRPC server get restarted, the client should be mostly intact, the underlying channel could be fall into an unavailable status, and depending on the client configuration, the error could be propagated to the user of the client or get silently retried.

Can you share your client configuration with more details? I'm new to the gRPC-Spring implementation, but this could lead to an area we can improve, with the default error handling & retry logic.

[ricardofernandess]

Have the same problem using both server and client implemented in Java. If the server has some temporary problem, when calling from the client I get io.grpc.StatusRuntimeException: UNAVAILABLE: io exception

Even if the server recovers from this failure i have to restart the client, otherwise the client gives the same error without calling the server again.

[ricardofernandess]

Upon further research it has to do with connection backoff described here https://github.com/grpc/grpc/blob/v1.32.0/doc/connection-backoff.md

Haven't found a way to disable this behaviour

[ST-DDT]

You might want to add a GrpcChannelConfigurer.
Although I'm not sure which option you need to change those. But you could ask that upstream or on StackOverflow.

[fengli79]

In general, we have 2 layers need to cooperate on temp server failures.

• On the channel level, when the underlying connection is broken, the channel will try to reconnect with that backoff strategy.
• At the same moment, before a new connection get established and be available to use, the channel will in unavailable status, the RPC on that channel would get unavailable error.

Things missing here is to have a solid RPC retry strategy, thus the underlying connection issue can be hidden from the application layer. Unavailable error is typically retryable, but the application still need to be designed in a way to carefully distinguish idempotent and non-idempotent RPCs to avoid unsafe retries.