Phase-4/5 investigation found the live stall is a recovered standby with a divergent uncommitted tail (former-primary shape) that never reconverges: loops on ae_reject_logmismatch, its control channel flaps ~15s, each flap spawns an unwinnable PreVote failover while a live primary exists. - Phase 4 (Standby Reconvergence): lead with divergent-tail truncate+catch-up (make the AE backtrack + follower-conflict-truncate fire for a former-primary tail, not just a short suffix), keep the applied/log_len healthy gate. - Phase 5 (Election Liveness): retarget from 'missing election-restart' (retry works) to spurious-failover suppression — reconnect+catch-up on transient RDHUP, don't storm unwinnable PreVotes while a current-term primary is reachable. 5.2 now _Depends 4.2_. |
||
|---|---|---|
| .. | ||
| settings | ||
| specs | ||
| steering | ||