Re: LP_DEAD hinting and not holding on to a buffer pin on leaf page (Was: [WIP] Zipfian distribution in pgbench)

On Thu, Jul 27, 2017 at 10:05 PM, Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> I really don't know if that would be worthwhile. It would certainly fix
> the regression shown in my test case, but that might not go far enough.
> I strongly suspect that there are more complicated workloads where
> LP_DEAD cleanup from SELECT statements matters, which is prevented by
> the LSN check thing, just because there are always other sessions that
> modify the page concurrently. This might be true of Alik's Zipfian test
> case, for example.

I haven't studied the test case, but I think as a general principle it
makes sense to be happy when someone comes up with an algorithm that
holds a lock for a shorter period of time (and buffer pins are a type
of lock). There are a number of places (fast-path locking, for
example, or vacuum skipping pinned heap pages) where we have
fast-paths that get taken most of the time and slow paths that get
used when concurrent activity happens; empirically, such things often
work out to a win. I think it's disturbing that this code seems to be
taking the slow-path (which, in context, means skipping LP_DEAD
cleanup) even there is no concurrent activity. That's hard to
justify. But the fact that it is taking the slow-path when there *is*
concurrent activity is harder to complain about. That might win or it
might lose; the non-concurrent case only loses.

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