pg_auto_failover v1.6.1 (July 7, 2021)
This release contains monitor schema changes, so the monitor extension gets
a version bump from 1.5 to 1.6, and this is the first release in the 1.6
series.
In this release we introduce a new state in the FSM: "dropped". This state
allows a node to realise it's been dropped from the monitor, and act
accordingly (mostly, stops cleanly and register it's been dropped).
This means that the command pg_autoctl drop node now only completes when
the node that is being dropped is still reachable. To drop a node that is
unreachable (e.g. machine died), you should now use the command pg_autoctl drop node --force.
Added
- Feature crash recovery before pg_rewind (#656)
- Allow building pg_auto_failover with Postgres 14. (#716)
- Track nodes current timeline on the monitor and use it during election. (#730)
- Implement drop-at-a-distance semantics. (#734)
- Add the reported timeline id to the monitor events table. (#753)
Changed
- Fix how many nodes need to report their LSN to perform a failover. (#707)
- Allow an old primary node (demoted/catchingup) to join another election. (#727)
- Have pg_autoctl drop node command wait until the node has been removed. (#748)
Fixed
- Fix/consider timeline when reaching secondary (#695)
- Install btree_gist when upgrade to >= 1.4, not just 1.4. (#714)
- Fix a race condition issue around health check updates. (#720)
- Not all the monitor code got the memo about nodeid being a bigint. (#729)
- Use sysctl(3) on BSD (#733)
- Fix transaction begin failure handling (#751)
- Fix a connection leak at SIGINT. (#759)