Re: vacuumlo issue

On Tue, Mar 20, 2012 at 11:50 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Josh Kupershmidt <schmiddy@gmail.com> writes:
>> On Tue, Mar 20, 2012 at 7:53 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> I'm not entirely convinced that that was a good idea. However, so far
>>> as vacuumlo is concerned, the only reason this is a problem is that
>>> vacuumlo goes out of its way to do all the large-object deletions in a
>>> single transaction. What's the point of that? It'd be useful to batch
>>> them, probably, rather than commit each deletion individually.  But the
>>> objects being deleted are by assumption unreferenced, so I see no
>>> correctness argument why they should need to go away all at once.
>
>> I think you are asking for this option:
>>   -l LIMIT     stop after removing LIMIT large objects
>> which was added in b69f2e36402aaa.
>
> Uh, no, actually that flag seems utterly brain-dead.  Who'd want to
> abandon the run after removing some arbitrary subset of the
> known-unreferenced large objects?  You'd just have to do all the search
> work over again.  What I'm thinking about is doing a COMMIT after every
> N large objects.
>
> I see that patch has not made it to any released versions yet.
> Is it too late to rethink the design?  I propose (a) redefining it
> as committing after every N objects, and (b) having a limit of 1000
> or so objects by default.

I'll dispute the characterization of "utterly brain-dead"; it's better
than what we had before, which was nothing.  However, I think your
proposal might be better still.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company