Re: Re: BUG #12990: Missing pg_multixact/members files (appears to have wrapped, then truncated)

Thomas Munro wrote:

> One thing I noticed about your patch is that it effectively halves the
> amount of multixact members you can have on disk. Sure, I'd rather
> hit an error at 2^31 members than a corrupt database at 2^32 members,
> but I wondered if we should try to allow the full range to be used.

Ah, yeah, we do want the full range; that's already built in the code
elsewhere.

In this version, I used your WouldWrap function, but there was a bug in
your formulation of the call site: after the WARNING has been issued
once, it is never issued again for that wraparound cycle, because the
second time around the nextOffset has already crossed the boundary and
your routine returns false. IMO this is wrong and the warning should be
issued every time. To fix that problem I removed the offsetWarnLimit
altogether, and instead do WouldWrap() of the value against
offsetStopLimit minus the 20 segments. That way, the warning is issued
continuously until the offsetStopLimit is reached (once there,
obviously, only the error is thrown, not the warning, which is correct.)

I also added a call to DetermineSafeOldestOffset() in TrimMultiXact:
as far as I can tell, this is necessary for the time when a standby
exits recovery, because when InRecovery we return early from
DetermineSafeOldestOffset() so the safe point would never get set.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services