hyperledger-labs / fabric-smart-client / 26579103782

74%

Build Type

push

github

Committed by web-flow

Commit Message

fix(fabric/ordering): always discard failed connections in BFT broadcast (#1485)

* fix(fabric/ordering): always discard failed connections in BFT broadcast

Broadcast() only ran its usedConnections cleanup on the overall-failure path. When a broadcast partially failed but still cleared the BFT success threshold -- e.g. 3 of 4 orderers acked, the 4th errored on Send/Recv or returned a non-SUCCESS status -- the failed connections never got discarded. Their connSem semaphore units stayed held and their gRPC streams stayed open for the lifetime of the process. Under sustained partial-failure conditions (one slow/flaky orderer that intermittently errors on Recv), this compounds until connSem is exhausted, after which getConnection falls back to its slow path and throughput collapses.

Move the cleanup loop above the threshold check so it runs on every path. Successful connections are already returned to their pools by releaseConnection (which does not touch the semaphore), so this only affects the previously-leaking failed-connection set.

Test: TestBFTBroadcaster_DiscardsFailedConnectionsOnPartialFailureSuccess exercises the success-with-partial-failure path against a pre-filled connection pool and asserts that exactly one semaphore unit is freed after the broadcast returns. Pre-fix the test fails (no unit released); post-fix it passes.

Signed-off-by: Evan <evanyan@sign.global>
Signed-off-by: Marcus Brandenburger <mbrandenburger@users.noreply.github.com>
Co-authored-by: Marcus Brandenburger <mbrandenburger@users.noreply.github.com>

Coverage Stats

4 of 4 new or added lines in 1 file covered. (100.0%)

8 existing lines in 4 files now uncovered.

28370 of 39847 relevant lines covered (71.2%)

2530.9 hits per line