Thread

Re: error: could not find pg_class tuple for index 2662

Tom Lane <tgl@sss.pgh.pa.us> — 2011-08-04T23:38:28Z

Ahh ... you know what, never mind about stack traces, let's just see if
the attached patch doesn't fix it.

I still haven't reproduced the behavior here, but I think I see what
must be happening: we are getting an sinval reset while attempting to
open pg_class_oid_index.  The latter condition causes its refcount to
be above 1, which will cause RelationClearRelation to directly call
RelationReloadIndexInfo, which enables the following sequence of calls:
RelationCacheInvalidate -> RelationClearRelation -> RelationReloadIndexInfo.
And the problem is that RelationCacheInvalidate intentionally forces
pg_class_oid_index to be updated first.  That was okay when the code was
written, because the relcache entry for pg_class itself never really
needed any updates.  Now it does, so we have to make sure pg_class gets
updated first, *then* pg_class_oid_index (which might result in a
seqscan of pg_class), then everything else (for which we'll try to use
pg_class_oid_index to locate their pg_class tuples).

			regards, tom lane