Nodes in the down state cannot failback

[feeling0713]

If there are unhealthy nodes in the database cluster, starting pgpool at this time will not initiate the health check for the corresponding failed nodes. Therefore, even if the subsequent databases return to normal, pgpool will still mark them as down. Why is this design chosen? Why does the health check only focus on healthy nodes?

[tatsuo-ishii]

Yes, it's by design. Problem is, there's no reliable way for pgpool to confirm the database returns to normal. Their data maybe outdated, or even the data is corrupted. Just succeeding in starting PostgreSQL is not reliable enough. The only reliable method is a manual intervention. Once a database administrator confirms the database is healthy, he/she can issue pcp_attach_node, and the health check will restart the checking.

If the database is not super critical, you might want to use auto_failback feature of pgpool. It's up to your decision.

[feeling0713]

However, the prerequisite for using auto_failback is that the health check process must exist; in this scenario, even if it is enabled, it will be ineffective