Re: pg_upgrade < 9.3 -> >=9.3 misses a step around multixacts

On 2014-07-21 12:20:56 -0400, Tom Lane wrote:
> Andres Freund  writes:
> > 2) From a pedantic POV lastSane* should be initialized to
> > ReadNewTransactionId()/ReadNextMultiXactId(). On a busy system we could
> > otherwise consider concurrently updated values as being too new if some
> > longrunning transaction finishes. Also GetOldestXmin() can sometimes go
> > backwards...
> 
> If it's wrong then that's a pre-existing bug, but I don't think it's
> wrong.

I'm not wondering so much about vac_update_relstats(). There indeed the
calls aren't new and the worst that can happen is a slightly older
freeze limit. I'm more wondering about vac_update_datfrozenxid(). Afaics
we very well could hit
	newFrozenXid = lastSaneFrozenXid = GetOldestXmin(NULL, true);
	newMinMulti = lastSaneMinMulti = GetOldestMultiXactId();
...
		/*
		 * If things are working properly, no relation should have a
		 * relfrozenxid or relminmxid that is "in the future".  However, such
		 * cases have been known to arise due to bugs in pg_upgrade.  If we
		 * see any entries that are "in the future", chicken out and don't do
		 * anything.  This ensures we won't truncate clog before those
		 * relations have been scanned and cleaned up.
		 */
		if (TransactionIdPrecedes(lastSaneFrozenXid, classForm->relfrozenxid) ||
			MultiXactIdPrecedes(lastSaneMinMulti, classForm->relminmxid))
		{
			bogus = true;
			break;
		}

for a relation that has just been vacuumed by another backend. In a
database with a high number of pg_class entries/much bloat it's not that
unrealistic that relfrozenxid/relminmxid have been updated since the
GetOldest* calls above. Since vac_update_relstats() writes them using
heap_inplace_update() the window for that isn't particularly small. That
could potentially lead to never updating the database freeze limits,
right?

Greetings,

Andres Freund

-- 
 Andres Freund	                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services