Re: [WIP] Keeping track of snapshots

Gregory Stark wrote:
> "Alvaro Herrera"  writes:
> 
> > The other question is about CommitTransactionCommand.  Currently my
> > EOXact routine barfs for every snapshot not unregistered on main
> > transaction commit -- a leak.  I see this as a good thing, however it
> > forced me to be more meticulous about not having ActiveSnapshot be set
> > in commands that have multiple transactions like VACUUM, multitable
> > CLUSTER and CREATE INDEX CONCURRENTLY.
> 
> I believe ActiveSnapshot has to be set during CREATE INDEX CONCURRENTLY if
> it's an expression index which calls a function which needs a snapshot...

Yeah, it is set at that point -- it is unset just before
CommitTransactionCommand is called.  Same for multitable CLUSTER -- it
certainly does need a snapshot, but it needs to unregister it just
before committing at the end of processing each table.

> AFAICT VACUUM had better not ever need a snapshot because its xmin isn't
> included in other vacuum commands' globalxmin so there's no guarantee that if
> it had a snapshot that the tuples visible in that snapshot wouldn't disappear
> out from under it.

Yeah, I neglected to mention that this is just for VACUUM FULL (which
does have its xmin included in snapshots).

-- 
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.