Expand description
FailoverWatchdog - scans the heartbeat table and reclaims
in-flight work whose owning process has stopped beating.
Architectural contract: failover happens within ONE epoch of the
peer’s last beat. The watchdog advances the global epoch on each
scan; any process whose last_seen_epoch < global - grace_epochs
is presumed dead and its in_flight_bitmap is returned to the
caller as a ReclaimReport so the caller (typically the
scheduler) can reassign the work.
Structs§
- Failover
Watchdog - Watchdog scanner; one per cooperating cluster.
- Reclaim
Report - Report from one watchdog scan.
Constants§
- DEFAULT_
GRACE_ EPOCHS - Default grace window. A slot must miss more than this many epochs before it is reclaimed.