Horton

Re: Failures in constraints regression test, "read only 0 of 8192 bytes"

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Heikki Linnakangas <hlinnaka@iki.fi>
Cc: Alexander Lakhin <exclusion@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, Tomas Vondra <tomas.vondra@enterprisedb.com>, Ronan Dunklau <ronan.dunklau@aiven.io>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2024-06-20T23:12:22Z
Lists: pgsql-hackers 
Heikki Linnakangas <hlinnaka@iki.fi> writes:
> In commit af0e7deb4a, I removed the call to RelationCloseSmgr() from 
> RelationCacheInvalidate(). I thought it was no longer needed, because we 
> no longer free the underlying SmgrRelation.

> However, it meant that if the relfilenode of the relation was changed, 
> the relation keeps pointing to the SMgrRelation of the old relfilenode. 
> So we still need the RelationCloseSmgr() call, in case the relfilenode 
> has changed.

Ouch.  How come we did not see this immediately in testing?  I'd have
thought surely such a bug would be exposed by any command that
rewrites a heap.

			regards, tom lane

Commits

  1. Fix relcache invalidation when relfilelocator is updated

  2. Don't destroy SMgrRelations at relcache invalidation

  3. Add a new slot sync worker to synchronize logical slots.