mirror of
https://github.com/stablyai/orca.git
synced 2026-09-22 16:02:32 +00:00
Jinwoo-H
ad8de92876
Two review findings on the persisted close reason. The clear set the fence back to NULL, so an unawaited record from an older close could land after a later proof, match IS NULL and stamp signed-out onto a host that had just proved itself. A proof now writes its own timestamp instead of erasing one, and one rule decides both writes: the newer event wins, ties to the close, because a close cannot precede the proof of the socket it closes. The clear also sat on the new-session path only. A control close inside the orphan grace leaves a session a reconnect resumes, and that rebind returns before ever reaching it. The clear now sits at the last point both admissions share.