Re: Rare corruption of pg_class index

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Tom Lane wrote:
> No, it's not a violation of ACID. In this case what you are doing is
> altering a table's schema without a sufficiently strong lock on the
> table, and that's a no-no, whether you would like it to be or not.

So as a general rule, the system tables should be considered a special
case as far as transactional activity? To be more precise, you are saying
that a system table must be locked in access exclusive mode before any
change is made to guarantee no problems occur?

> Well, ENABLE/DISABLE TRIGGER can't conceivably operate correctly without
> locking out writes, because it wouldn't be clear whether any particular
> write operation should fire the trigger or not. A hypothetical
> ENABLE/DISABLE RULE would be worse: AFAICS it'd have to lock out reads
> too, else it wouldn't be clear whether SELECTs should notice an ON
> SELECT rule. You can't get around those restrictions by trying to
> implement the enable/disable yourself via UPDATE pg_class; as you've
> found out, it just doesn't work.

So the oft-given advice of "UPDATE pg_class SET relhasrules = false"
is actually completely unsafe unless the entire referenced table is
completely locked, and unless you are using at least 8.2?

- --
Greg Sabino Mullane greg(at)turnstep(dot)com
End Point Corporation
PGP Key: 0x14964AC8 200701031037
http://biglumber.com/x/web?pk=2529DF6AB8F79407E94445B4BC9B906714964AC8
-----BEGIN PGP SIGNATURE-----

iD8DBQFFm83kvJuQZxSWSsgRAvhiAJ0TEgzg3mkZ6BKvlR7jHKY96JVmmACgxrYb
pFY851yrZd5bh7GHaR+k0Q4=
=FgL+
-----END PGP SIGNATURE-----