Re: Removing freelist (was Re: Should I implement DROP INDEX CONCURRENTLY?)

Tom Lane <tgl@sss.pgh.pa.us>

From: Tom Lane <tgl@sss.pgh.pa.us>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Jim Nasby <jim@nasby.net>, Heikki Linnakangas <heikki.linnakangas@enterprisedb.com>, Simon Riggs <simon@2ndquadrant.com>, Daniel Farina <daniel@heroku.com>, Merlin Moncure <mmoncure@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2012-01-23T16:01:52Z
Lists: pgsql-hackers
Robert Haas <robertmhaas@gmail.com> writes:
> On Mon, Jan 23, 2012 at 12:12 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> The expensive part of what
>>> we do while holding BufFreelistLock is, I think, iterating through
>>> buffers taking and releasing a spinlock on each one (!).

>> Yeah ... spinlocks that, by definition, will be uncontested.

> What makes you think that they are uncontested?

Ah, never mind.  I was thinking that we'd only be touching buffers that
were *on* the freelist, but of course this is incorrect.  The real
problem there is that BufFreelistLock is also used to protect the
clock sweep pointer.  I think basically we gotta find a way to allow
multiple backends to run clock sweeps concurrently.  Or else fix
things so that the freelist never (well, hardly ever) runs dry.

			regards, tom lane