hyperledger-labs / fabric-smart-client / 37603115914

91%

Build Type

push

github

Committed by web-flow

Commit Message

fix(view): dial outside provider lock and set websocket handshake timeout (#2036)

MultiplexedProvider.NewClientStream previously held the provider lock (p.mu)
across both the network dial to the remote peer and the subsequent sub-connection
creation (newClientSubConn), which writes the meta frame with a 2-minute write
timeout. Additionally, OpenWSClientConn created a websocket.Dialer with zero
HandshakeTimeout, causing dials to unresponsive peers to block indefinitely.
Together, these issues allowed a single slow or unresponsive peer to block
stream creation to every other peer on the node.

Address both issues:
- Move websocket dial and newClientSubConn execution outside p.mu.
- Deduplicate concurrent dials to the same peer/URL using per-URL in-flight
  tracking (inFlightDial) so concurrent callers continue to share a single
  connection without blocking stream creation to other peers.
- Set DefaultHandshakeTimeout (30s) on gorilla/websocket dialer in
  OpenWSClientConn, and provide OpenWSClientConnContext for context-bounded
  dials with dialer.DialContext.
- Document on MultiplexedProvider.mu that it is never held across a network
  dial or sub-connection meta write.
- Add regression tests verifying that a hanging peer does not block stream
  creation to healthy peers, concurrent dials share the connection,
  unresponsive handshakes time out cleanly, stale onClose callbacks do not evict
  reconnected clients, and post-shutdown stream creations are rejected.

Fixes #1975

Signed-off-by: Surbhi Agarwal <SurbhiAgarwal1@users.noreply.github.com>
Co-authored-by: Surbhi Agarwal <SurbhiAgarwal1@users.noreply.github.com>

Coverage Stats

77 of 92 new or added lines in 2 files covered. (83.7%)

11 existing lines in 5 files now uncovered.

29018 of 32003 relevant lines covered (90.67%)

2782.96 hits per line