Client can hang when connecting to Docker port-forwarded gRPC server that goes offline

[Obbut]

We're experiencing an issue when running a gRPC server in Docker, using port forwarding. When a client connects (like an iOS app or another executable on the same system) using the forwarded port, initially all goes well. However, if we stop the server for a short time (say, 10 seconds) and then restart it, while still trying to call RPCs from the client, the client hangs for at least a few minutes without cancellation, errors, or trying to re-establish the connection.

It's perhaps easier to see for yourself, so I've created reproduction steps:

• Clone https://github.com/Obbut/GRPCNIOTransportIssue2
• Open the cloned repo in VS Code and launch the dev container. You need Docker and the Dev containers extension for this. If I remember correctly, the dev containers extension is installed by default.
• Run both the server and the client and wait until the client stalls. The client can run from the same container because it connects to host.docker.internal

[glbrntt]

Thanks for reporting this and providing a repro, I'll try to take a look today or tomorrow.

[glbrntt]

hey @Obbut apologies for not getting back to you sooner, I've been having issues setting up the repro (related to VSCode and dev containers, I think a setup issue on my side but I haven't had time to dig into).

Should this reproduce if I just run the server in Docker and run the client locally?

[Obbut]

I think it should reproduce if you run the server in Docker and the client locally, as long as you're connecting to the server through a Docker port forward. I didn't test it that way though

[glbrntt]

I couldn't reproduce this by running the server in Docker. Although I noticed the server was listening on 127.0.0.1 (client couldn't connect at all) which I changed to 0.0.0.0 (no issues reconnecting).

[Obbut]

Thanks for getting back on this. I'll make some time next week to provide better reproduction steps.