hyperledger-labs / fabric-smart-client / 41068046242

92%

Build Type

push

github

Committed by web-flow

Commit Message

fix(view): reply to unknown views without loops (#2415)

A first P2P message for a view the node has no responder for is answered with the error `unknown_view: <view>`, so the initiator fails at once instead of waiting a minute for its session to time out. This is the usual failure between two versions of an application, or after a view moved to another package.

Every error reply a node sends unasked goes through one reject path in the view P2P service. Its payload is `<reason>: <detail>`, the detail reduced to printable ASCII and 256 bytes. The busy reply reads `busy: too many concurrent responders for this peer`. Rejections share the per-peer bound on pending rejections and are counted by `fsc_view_services_view_p2p_rejections{reason}`, wired in through the new additive `p2p.WithMetrics` option.

Two guards keep rejections cheap and finite:
- A first message with status ERROR or an empty caller is dropped. Rejections are such messages, so two nodes cannot reject each other in a loop, and an ERROR message no longer starts a responder.
- `ReplyError` writes only to a stream to the peer that already exists and never dials. Without one, the rejection is dropped.

The pingpong integration suite gains an `unhosted` view that only the initiator registers. Calling it fails with `unknown_view: ...` on libp2p and WebSocket.

docs/upgrade/v0.26-to-v0.27.md is the operator and developer upgrade guide for every change since v0.26.0, including this one.

Fixes #2409

Signed-off-by: Marcus Brandenburger <bur@zurich.ibm.com>

Coverage Stats

53 of 53 new or added lines in 3 files covered. (100.0%)

8 existing lines in 4 files now uncovered.

30131 of 32933 relevant lines covered (91.49%)

1260.22 hits per line