Timeline-fork detection: fix #683 unrecoverable state after failed switchover - #1161

Merged

dimitri merged 28 commits into

mainfrom

timeline-fork-detection

Jul 27, 2026

Commits on Jul 27, 2026

Part 1: verify genuine timeline ancestry at the secondary gate, not just numbers
dimitri
committed
60ce50e
Part 2 (schema + publishing): node_timeline_history table and reporting
dimitri
committed
2c8ec55
Part 2 (election): filter report_lsn candidates by timeline ancestry
dimitri
committed
2a04ae9
Part 3: operator-pinned accepted_timeline for genuine forks
dimitri
committed
b1e9263
Part 4: pg_autoctl show timeline
dimitri
committed
52cb03c
Add monitor regression tests for timeline-fork detection
dimitri
committed
8ba911d
Fix node_timeline_history FK to cascade on node removal
dimitri
committed
9072320
Fix #683: standby ahead of upstream retried a doomed reconnect forever
dimitri
committed
303d30e
timeline_fork_report_lsn_deadlock.pgaf: add live-cluster show/accept timeline coverage
dimitri
committed
32526b6
ci: add every .pgaf spec to a CI schedule
dimitri
committed
29f8e65
fast_forward.pgaf: fix test_003 deadlock by calling perform_failover via SQL
dimitri
committed
af9cb48
Drop fast_forward from this branch's CI schedule and revert its spec
dimitri
committed
187ee3a
Fix citus_indent formatting in timeline_history.c/.h
dimitri
committed
1c61489
monitor.c: fix show timeline fork message to not assert election behavior
dimitri
committed
658d269
docs: document timeline-fork detection (accept/show timeline, FSM, fault-tolerance, architecture, FAQ, operations)
dimitri
committed
495ac43
monitor: add missing SQL objects to the 2.2->2.3 upgrade script
dimitri
committed
17d0ca4
docs+tests: address review feedback on timeline-fork-detection docs
dimitri
committed
69f35e8
pgaftest: size expect-mismatch error buffers to fit multi-row output
dimitri
committed
56d0ee8
docs+tests: show concurrent application traffic during a 3-node fork
dimitri
committed
e21947f
monitor: detect a timeline fork on a secondary as soon as it's reported
dimitri
committed
7037c99
pgctl: pg_rewind must connect to the formation's own database, not a hardcoded postgres
dimitri
committed
365bb62
docs: describe the new automatic, near-instant catchingup push on a forked secondary
dimitri
committed
79c3a3d
timeline_fork_report_lsn_deadlock.pgaf: fix race between the new auto-catchingup guard and the maintenance-cycle step
dimitri
committed
e709c8b
fast_forward: fix silent promotion corruption in multi-node failover
dimitri
committed
58c0b7d
primary_standby: detect same-timeline LSN divergence in standby_check_timeline_with_upstream
dimitri
committed
c5846f6
fsm_transition_citus: fix same fast_forward promotion corruption for Citus workers
dimitri
committed
13b7984
fsm_mermaid.c: fix citus_indent formatting (missing blank line)
dimitri
committed
0f70556
tests: use SIGINT for cluster teardown, stop leaking orphaned nodes
dimitri
committed
5e8a73f