test linger enabled - resident reuse a resume inside the TTL reuses the SAME live pid (subagent_linger_wake_test.exs:119) failed on main run 32996283175: pid2 != pid1 with a 60s TTL, meaning the resident Loop was terminated and replaced between the two run_subagent calls despite linger.
Not an env race: only this file sets :subagent_linger_ms and it is async: false. Does not reproduce locally (8 consecutive green runs of the file). Load-dependent - something under CI contention kills or replaces the lingered Loop between completion and resume (an internal idle/inactivity path? registry handoff race in the fast-wake path?).
Leaving to the orchestrator owner rather than guessing at product code; a rerun of the failed job on the same SHA is in flight to confirm flakiness. Two other main flakes fixed this week for reference: agent_runtime_routes cleanup race (#178) and the PTY version-mismatch banner (#176).
test linger enabled - resident reuse a resume inside the TTL reuses the SAME live pid(subagent_linger_wake_test.exs:119) failed on main run 32996283175:pid2 != pid1with a 60s TTL, meaning the resident Loop was terminated and replaced between the tworun_subagentcalls despite linger.Not an env race: only this file sets
:subagent_linger_msand it isasync: false. Does not reproduce locally (8 consecutive green runs of the file). Load-dependent - something under CI contention kills or replaces the lingered Loop between completion and resume (an internal idle/inactivity path? registry handoff race in the fast-wake path?).Leaving to the orchestrator owner rather than guessing at product code; a rerun of the failed job on the same SHA is in flight to confirm flakiness. Two other main flakes fixed this week for reference: agent_runtime_routes cleanup race (#178) and the PTY version-mismatch banner (#176).