On 23 October 2013 02:18, Andres Freund <andres@2ndquadrant.com> wrote:
> Hi,
>
> Using the same debugging hack^Wpatch (0001) as in the matview patch
> (0002) an hour or so ago I noticed that INSERT INTO view WITH CHECK
> doesn't lock the underlying relations properly.
>
> I've attached a sort-of-working (0003) hack but I really doubt it's the
> correct approach, I don't really know enough about that area of the
> code.
> This looks like something that needs to be fixed.
>
Hmm, my first thought is that rewriteTargetView() should be calling
AcquireRewriteLocks() on viewquery, before doing too much with it.
There may be sub-queries in viewquery's quals (and also now in its
targetlist) and I don't think the relations referred to by those
sub-queries are getting locked.
I think that any code that is doing anything significant with a rule
action's query needs to think about locking the query's relations. I
did a quick search and the only suspicious code I found was the
matview and auto-updatable view code.
Regards,
Dean
> Also attached is 0004 which just adds a heap_lock() around a newly
> created temporary table in the matview code which shouldn't be required
> for correctness but gives warm and fuzzy feelings as well as less
> debugging noise.
>
> Wouldn't it be a good idea to tack such WARNINGs (in a proper and clean
> form) to index_open (checking the underlying relation is locked),
> relation_open(..., NoLock) (checking the relation has previously been
> locked) and maybe RelationIdGetRelation() when cassert is enabled? ISTM
> we frequently had bugs around this.
>
> Greetings,
>
> Andres Freund
>
> --
> Andres Freund http://www.2ndQuadrant.com/
> PostgreSQL Development, 24x7 Support, Training & Services
>
>
> --
> Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-hackers
