Icinga alerts

PostgreSQL: replication too far behind

“replication on the PostgreSQL slave is too far behind master [value in bytes]”

When this alert fires, the Postgres slave replication process may be struggling to keep up (due to load) or may have stopped altogether. Take a look at the graph linked by Icinga historical information.

If one node is down (resulting in Graphite having NULL values for one metric) then the reported lag will jump to size of the XLOG position on the remaining node and raise a CRITICAL alert.

If both nodes are down then Graphite will return no data and an UNKNOWN alert will be raised. This should correlate with other checks for server and database health.

If a backup has recently been restored to the primary, it may be that the primary did not keep enough WAL segments around to bring the standby up to date. This can be fixed by restore the backup to the standby as well, after which normal replication will take over.

You can get a quick view of postgresql’s wal by doing the following:

on a postgresql-primary: ps -ef | grep sender

on a postgresql-standby: ps -ef | grep receiver

If the slave has fallen too far behind or is in an otherwise unrecoverable state then you may need to resync it.

If this is a new machine, or if you have recently resynced it, the issue may be that PostgreSQL replication is set up and running correctly, but collectd needs to be restarted on either the primary or standby to ensure that Graphite is receiving the metrics it needs.

$ sudo service collectd restart

The replication lag is measured by examining the difference in the XLOG location in bytes.