fix: catch discard error and return

[ningmingxiao]

to fix containerd/containerd#12669 @fuweid @dmcgowan

func (c *serverConn) run(sctx context.Context) {
....
	go func(recvErr chan error) {
		defer close(recvErr)
		for {
			select {
			case <-c.shutdown:
				return
			case <-done:
				return
			default: // proceed
			}

			mh, p, err := ch.recv()
			if err != nil {
				status, ok := status.FromError(err)
				if !ok {
					recvErr <- err
					return
				}

				// in this case, we send an error for that particular message
				// when the status is defined.
				if !sendStatus(mh.StreamID, status) {
					return
				}

				continue
			}

recvErr will not receive an error when discard failed and go func will not return.

or we can catch discard error and return

func (c *serverConn) run(sctx context.Context) {
...
		case err := <-recvErr:
			// TODO(stevvooe): Not wildly clear what we should do in this
			// branch. Basically, it means that we are no longer receiving
			// requests due to a terminal error.
			recvErr = nil // connection is now "closing"
			if errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF) || errors.Is(err, syscall.ECONNRESET) {
				// The client went away and we should stop processing
				// requests, so that the client connection is closed
				return
			}
			log.G(ctx).WithError(err).Error("error receiving message")
                      //catch error here 
			_, ok := err.(syscall.Errno)
			if ok {
				return
			}

[kzys]

Sorry for the late reply. I've updated GitHub Actions. Can you rebase this PR against main?

[ningmingxiao]

done thanks @kzys

[ningmingxiao]

can you review this pr @kzys @dmcgowan

[dmcgowan]

Did you confirm this fixed the issue? Looking through the bufio implementation of Discard, it is already doing something similar on the existing buffer. I don't see the allocation in that implementation either, it seems to just iterate filling up the existing buffer (defaulted to 4k).

[ningmingxiao]

you are right I create new pr will fix. @dmcgowan

[dmcgowan]

Catching any error on discard makes sense and exiting. If the discard did not succeed then the connection must be ended as it is no longer in a processable state.

[ningmingxiao]

done thanks @dmcgowan