prisma-risk / tsoracle / 26367973536

95%

push

github

fix(server): surface tonic drain errors when the watch arm fires (#293)

serve_with_shutdown and serve_with_listener discarded tonic's graceful-drain result via `let _ = serve.await` when the leader-watch task terminated first. A drain failure (port stolen, resource exhaustion) was hidden behind a clean-shutdown report whenever the watch itself ended cleanly.

Add combine_watch_and_drain, which prefers the watch error when both fail (poisoned serving state was already published, so it is the root cause) and surfaces a drain error only when the watch outcome is otherwise Ok. Both serve arms route through it. The helper is generic over the drain error type so the precedence logic is unit-testable without fabricating a tonic::transport::Error (which has no public constructor): production passes the real transport error, tests pass a synthetic ServerError via the reflexive From impl.

Closes #243

53 of 55 new or added lines in 1 file covered. (96.36%)

11047 of 11653 relevant lines covered (94.8%)

432067.02 hits per line