>>>>> "Andrew" == Andrew Gierth <andrew@tao11.riddles.org.uk> writes:
Tom> I have not looked very closely, but I'm suspicious that the test
Tom> case depends on no autovacuum transactions running concurrently
Tom> with it. Disabling autovac on the table itself is not enough to
Tom> control whether global xmin is being held back by some other
Tom> autovac transaction somewhere.
Andrew> Aren't vacuum transactions ignored for that?
Andrew> vacuum_set_xid_limits is certainly passing ignoreVacuum=true to
Andrew> GetOldestXmin.
Answering my own question: vacuum transactions are ignored but analyze
transactions aren't, and autovacuum will happily kick off an autoanalyze
of pg_attribute during the test.
Simplest fix seems to just be to remove the delete statement; I've
verified that the problem can be reproduced without it, and that should
make the result independent of OldestXmin. I'll commit that shortly
unless I hear to the contrary.
--
Andrew.