Post-Phase-2 chaotic wedge root-caused from frozen soak logs: primary stable at term 6, but recovered standby ctrl-2 (divergent uncommitted tail 12-15) cannot reconverge; its primary control channel RDHUP-flaps ~15s, each flap spawns an unwinnable PreVote failover (peers see primary alive, no grant) -> timeout -> reconnect -> repeat, so the cluster never reaches stable 3-of-3. Fix spans Phase 4 (truncate divergent tail + catch up like ctrl-3 does) and Phase 5 (don't launch unwinnable failover storm on transient RDHUP while a current-term primary is reachable; reconnect+catch-up first). RED direction recorded. (Dispatched investigator hung; reconstructed by parent.) |
||
|---|---|---|
| .. | ||
| settings | ||
| specs | ||
| steering | ||