CI: end-to-end connector tests time out when the runner host is saturated #78

Open
opened 2026-09-04 18:21:40 +09:00 by natsukium · 0 comments
Owner

Push-event run 1646 on main (03d56ca7, a BREAKING.md-only change) failed in cargo nextest: connector::tests::window_client_hears_session_exited_when_shell_exits hit its 15 s deadline (crates/felis-client-core/src/connector.rs:2844) and reported left: None. The same SHA passed the pull_request run minutes earlier. The runner is the host that also builds the implementation workflows; its one-minute load average was around 42 on 12 cores at the time. The Criterion regression gate reported phantom +17% to +51% regressions on a comment-only PR (#76) under the same load. The bench gate now withholds its verdict on a saturated host; the deadline-based daemon tests have no such guard. Options: scale the deadlines by load, serialize the test job behind other runner work, or accept the flake and rerun. Not a code regression.

Push-event run 1646 on main (03d56ca7, a BREAKING.md-only change) failed in cargo nextest: `connector::tests::window_client_hears_session_exited_when_shell_exits` hit its 15 s deadline (`crates/felis-client-core/src/connector.rs:2844`) and reported `left: None`. The same SHA passed the pull_request run minutes earlier. The runner is the host that also builds the implementation workflows; its one-minute load average was around 42 on 12 cores at the time. The Criterion regression gate reported phantom +17% to +51% regressions on a comment-only PR (#76) under the same load. The bench gate now withholds its verdict on a saturated host; the deadline-based daemon tests have no such guard. Options: scale the deadlines by load, serialize the test job behind other runner work, or accept the flake and rerun. Not a code regression.
Sign in to join this conversation.
No description provided.