Improve error reporting [#3595]

pkg/api/v3/message/transport.go

@@ -291,10 +291,23 @@ func (c *RoutedTransport) dial(ctx context.Context, addr multiaddr.Multiaddr, st
 			return s, nil // Success
 		}
 
-		// Return the error if it's a client error (e.g. misdial)
 		var err2 *errors.Error
 		if errors.As(err, &err2) && err2.Code.IsClientError() {
-			return nil, errors.UnknownError.Wrap(err)
+			switch {
+			case err2.Code.IsClientError():
+				// Return the error if it's a client error (e.g. misdial)
+				return nil, errors.UnknownError.Wrap(err)
+
+			case err2.Code == errors.PeerMisbehaved:
+				// Find the cause of the misbehavior report
+				for err2.Cause != nil && err2.Code == errors.PeerMisbehaved {
+					err2 = err2.Cause
+				}
+
+				// If the error is an encoding issue, log it and return "internal error"
+				multi.BadDial(ctx, addr, s, err)
+				return nil, errors.InternalError.WithFormat("internal error: %w", err)
+			}
 		}
 
 		// Remove the stream from the dictionary

pkg/api/v3/p2p/dial/dialer.go

@@ -100,9 +100,19 @@ func (d *dialer) Dial(ctx context.Context, addr multiaddr.Multiaddr) (stream mes
 func (d *dialer) BadDial(ctx context.Context, addr multiaddr.Multiaddr, s message.Stream, err error) bool {
 	ss, ok := s.(*stream)
 	if !ok {
+		slog.InfoCtx(ctx, "Bad dial", "address", addr, "error", err)
 		return false
 	}
+
 	slog.InfoCtx(ctx, "Bad dial", "peer", ss.peer, "address", addr, "error", err)
+
+	var err2 *errors.Error
+	if errors.As(err, &err2) && err2.Code == errors.EncodingError {
+		// Don't mark a peer bad if there's an encoding failure. Is this a good
+		// idea?
+		return false
+	}
+
 	d.tracker.Mark(ss.peer, addr, api.PeerStatusIsKnownBad)
 	return true
 }

scripts/delve-in-docker.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+SCRIPT="dlv attach 1 --headless --listen=:2345 --accept-multiclient --api-version=2"
+
+docker exec --privileged -it "$1" bash -c "$SCRIPT"