Re: out-of-order XID insertion in KnownAssignedXids

Hi,

On 2016-10-18 14:57:52 +0200, fredrik(at)huitfeldt(dot)com wrote:
> we are running many postgresql master/slave setups. The slaves are
> initialised from a pg_basebackup from the master and are sync
> streaming from the master. When we determine the master has failed,
> the slave is promoted. Some time after that, the old master is again
> initialised with a pg_basebackup and starts streaming from the new
> master.

Could you describe in a bit more detail how exactly you're setting up
the standbys? E.g. the exact recovery.conf used, whether you remove any
files during starting a standby. Also how exactly you're promoting
standbys?

> Recently, we have gotten this error a fair amount of times: "out-of-order XID insertion in KnownAssignedXids" when postgresql attempts to start after being initialised with a pg_basebackup from the current master.

Which version are you encountering this on precisely?

> Once the issue has occurred, a subsequent re-initialisation (with a completely new pg_basebackup) does not resolve the issue.

How have you recovered from this so far?

> I have a setup in the failing state, so I can produce any kind of log mesages / details that would be helpful.

Could you use pg_xlogdump to dump the WAL file on which replay failed?
And then attach the output in a compressed manner?

Greetings,

Andres Freund