Users may observe PostgreSQL replication failures and repeated restarts of the automate-ha-postgresql service on follower nodes. The logs show pg_basebackup errors and child process exits with code 1, indicating replication setup or streaming issues.

Additionally, even when the wal replication timeouts are corrected, better performance is on average available by changing settings related to max_wal_size and checkpointing.

The issue occurs because the default values for wal_sender_timeout and wal_receiver_timeout are incorrectly configured to 60 milliseconds instead of 60 seconds. This causes premature disconnections between primary and standby nodes during replication attempts.

Additionally, excess load is placed on the system at its rated load levels when the settings for max_wal_size and checkpointing are not increased from their defaults.

1. Verify Current Settings:
   Log in to PostgreSQL and check the current timeout values:

   SHOW wal_sender_timeout;
   SHOW wal_receiver_timeout;
   

2. Temporarily Correct the Settings:
   Apply the following commands to adjust the timeouts at the database level:

   ALTER SYSTEM SET wal_sender_timeout = '60s';
   ALTER SYSTEM SET wal_receiver_timeout = '60s';
   SELECT pg_reload_conf();
   

   Monitor the system for stability before applying permanent configuration changes.

3. Apply Permanent Configuration Patch:
   Once stability is confirmed, patch the configuration starting with the leader node.
   Example configuration patch:

   [replication]
   wal_sender_timeout = '300s'
   wal_receiver_timeout = '120s'
   
   [postgresql]
   [postgresql.v1]
   [postgresql.v1.sys]
   [postgresql.v1.sys.pg]
   max_wal_size = "4GB"
   checkpoint_timeout = "15min"
   checkpoint_completion_target = 0.9
   

   This configuration reduces replication lag and improves disk performance under typical heavy load.

4. Restart Considerations:
   Be aware that applying configuration patches may trigger a leader election and restart of the PostgreSQL service, so configuration patching is best done in a quiet time or maintenance window.