UNAVAILABLE: Keepalive failed. The connection is likely gone

Question

I am using grpc-okhttp on an Android application for RPC calls to a backend.

This is the client side keep-alive configuration:

.keepAliveTime(2, TimeUnit.SECONDS)
.keepAliveTimeout(5, TimeUnit.SECONDS)
.keepAliveWithoutCalls(true)

I have observed cases where the connection dies at some point but is never restored until app restart.

I have not found a way to consistently reproduce it yet but in production I'm seeing a lot of errors "UNAVAILABLE: Keepalive failed. The connection is likely gone".

From my understanding gRPC should reconnect automatically but it looks like once the keep-alive fails once, it never even bothers to reconnect - following requests fail immediately, not even waiting for a deadline or keep-alive timeout.

Answer 1

This is likely caused by delayed/failed discovery of network state changes on Android devices. gRPC provides AndroidChannelBuilder, which is trying to address this problem particularly. It uses Android ConnectivityManager to receive network state updates and can respond more quickly to network changes.