Re: Bulk Insert tuning

On Tue, 2008-02-26 at 21:36 +0000, Simon Riggs wrote:
> On Tue, 2008-02-26 at 15:12 -0500, Tom Lane wrote:
> > Simon Riggs <simon@2ndquadrant.com> writes:
> > > Following patch implements a simple mechanism to keep a buffer pinned
> > > while we are bulk loading.
> >
> > This will fail to clean up nicely after a subtransaction abort, no?
>
> Yes, will fix.

Additional line in AbortSubTransaction handles this.

> > (For that matter I don't think it's right even for a top-level abort.)
> > And I'm pretty sure it will trash your table entirely if someone
> > inserts into another relation while a bulk insert is happening.
> > (Not at all impossible, think of triggers for instance.)
>
> The pinned buffer is separate from the preferred block for each
> relation; BulkInsertBuffer isn't used for determining the block to
> insert into. If you try to insert into a block that differs from the
> pinned one it unpins it and re-pins the new one. So it is always safe
> with respect to the data in the table.
>
> It can run into recursive bulk insert ops but that just destroys the
> performance advantage, its not actually dangerous.

I'm about to start refactoring code as suggested, so wanted to drop off
another version to allow everybody to examine the safety/not of this
approach. (So this patch is WIP)

--
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com

  PostgreSQL UK 2008 Conference: http://www.postgresql.org.uk