A stale suspicion timer could declare a node dead even after it had been refuted back to Alive via a higher incarnation number. check_suspicion_timeouts now verifies the node is still Suspect before calling declare_dead, and cancels the probe phase if the dead node was the current target. Also fix the 3-node gossip loop in node_death_tombstones_entries to deliver responses with the correct sender_id — previously all responses were attributed to a single hardcoded sender, causing false suspicions that triggered the protocol bug. Authored by Claude, lovingly guided by Zachery Aaron Shores-Chmielewski |
||
|---|---|---|
| .. | ||
| bin-runner | ||
| distribution | ||
| python | ||
| runtime-dashboard | ||
| simulation | ||
| std | ||
| wasm | ||