Re: Segmentation fault in 7.3 while vacuuming

I said:
> Frank van Vugt <ftm(dot)van(dot)vugt(at)foxi(dot)nl> writes:
>> ** while using a client written in TrollTech's QT v3.1 that creates a number
>> of temporary tables, uses a number lot of 'copy' statements and at the end
>> tries to 'vacuum full analyse' the database

> Drat --- looks like the local buffer manager code has gotten out of sync
> with the relcache code.

Here is the fix if you need to patch this locally.

regards, tom lane

*** src/backend/storage/buffer/localbuf.c.orig Wed Sep 4 16:31:25 2002
--- src/backend/storage/buffer/localbuf.c Thu Dec 5 17:48:10 2002
***************
*** 90,108 ****
{
Relation bufrel = RelationNodeCacheGetRelation(bufHdr->tag.rnode);

- /*
- * The relcache is not supposed to throw away temp rels, so this
- * should always succeed.
- */
- Assert(bufrel != NULL);
-
/* flush this page */
! smgrwrite(DEFAULT_SMGR, bufrel, bufHdr->tag.blockNum,
! (char *) MAKE_PTR(bufHdr->data));
! LocalBufferFlushCount++;

! /* drop refcount incremented by RelationNodeCacheGetRelation */
! RelationDecrementReferenceCount(bufrel);
}

/*
--- 90,113 ----
{
Relation bufrel = RelationNodeCacheGetRelation(bufHdr->tag.rnode);

/* flush this page */
! if (bufrel == (Relation) NULL)
! {
! smgrblindwrt(DEFAULT_SMGR,
! bufHdr->tag.rnode,
! bufHdr->tag.blockNum,
! (char *) MAKE_PTR(bufHdr->data));
! }
! else
! {
! smgrwrite(DEFAULT_SMGR, bufrel,
! bufHdr->tag.blockNum,
! (char *) MAKE_PTR(bufHdr->data));
! /* drop refcount incremented by RelationNodeCacheGetRelation */
! RelationDecrementReferenceCount(bufrel);
! }

! LocalBufferFlushCount++;
}

/*