The monitor doesn't update the nodes' states when replication connection is corrupted.

[gfphoenix78]

Version: 1.4.0

If we inject a network corruption between the primary and secondary, the monitor doesn't update the nodes' state.
So, the primary still tries to synchronize its WAL records to the secondary, which causes almost all queries blocked.

The monitor should be able to detect this scenario, update the nodes' states, and tell the primary to update its synchronized standbys.

[dimitri]

Fixed by #1149 (pgautofailover.replication_stall_timeout): the monitor now detects when a primary's own report shows no standby in pg_stat_replication and, after the configured timeout (10s default), assigns wait_primary to clear synchronous_standby_names and unblock writes — exactly this scenario, where both nodes stay reachable from the monitor but the primary-to-standby replication link itself is broken.