-
Type:
Problem report
-
Resolution: Unresolved
-
Priority:
Trivial
-
None
-
Affects Version/s: 8.0.0beta1 (master)
-
Component/s: Proxy (P), Server (S)
During testing, it was observed that internal server processes in Zabbix behave inconsistently when reconnecting to the database after a failover.
When a database node switch occurs, some internal processes produce the following sequence of logs:
- PGRES_FATAL_ERROR (from the old database connection)
- successful reconnection to another database node
- "database connection re-established"
At the same time, other internal processes show a different behavior:
- PGRES_FATAL_ERROR is logged
- reconnection to another database node happens successfully
- no "database connection re-established" log is produced
As a result, from an user perspective, it may look like some processes failed or did not recover correctly, even though they continue working normally.
Steps to reproduce:
- Configure a Zabbix Server with database failover (e.g. PostgreSQL primary/replica).
- Start the server and ensure all internal processes are running.
- Trigger a database node switch.
- Observe server logs during and after the failover.
Result:
- Some internal processes log "database connection re-established" after reconnection.
- Other processes reconnect successfully but do not log this message.
- All processes may log PGRES_FATAL_ERROR originating from the previous database node.
Expected: Logging of successful database reconnection should be consistent across all internal server processes. If a process logs a database connection error related to a lost connection and then successfully reconnects to another database node, the successful reconnection should be logged in