Doug Fawley

If the connection is lost in the middle of an RPC then it's expected to get RPC failures. You could use a `MaxAgeGrace` setting to help work around that issue...

How long after the initial GOAWAY are you seeing those EOF errors? Are you still pretty sure it's related? Does it still only reproduce every ~48 hours.

Let's leave this open. `end2end_test.go` is now over 7500 lines and still needs a cleanup.

It seems this still needs to be rebased due to outlierdetection files.

> we end up closing the stream with a http2.ErrCodeFrameSize This doesn't sound right, as that error is intended for violations of SETTINGS_MAX_FRAME_SIZE. > I couldn't figure out if this...

> Am I reading this wrong? I wasn't disagreeing with your assessment. I was stating that the behavior was incorrect AIUI. > No, we return INTERNAL from here. Hmm, that...

> Can we assume the http2 package will make sure that the header block is completely processed even when the limit is exceeded? I've been trying to verify that, but...

@dzhang-b that sounds like a gRPC-C++ issue; this is the grpc-go repo. You may want to [create an issue in the grpc/grpc repo](https://github.com/grpc/grpc/issues/new/choose) instead.

@menghanl was #5054 enough to resolve this issue or is there more to be done?

Maybe something like `grpc.ErrConnectionBroken` should be used for this kind of case. Regardless, the server side should not see RPC status errors at all, so Internal or any other code...