Re: GUC for cleanup indexes threshold.

Hi,

At Tue, 19 Sep 2017 16:55:38 -0700, Peter Geoghegan <pg@bowt.ie> wrote in <CAH2-Wzn0-3zxGRp_qp1OaEXY7h1W0-W_VCFO0nDv0K_+kabyYQ@mail.gmail.com>
> On Tue, Sep 19, 2017 at 4:47 PM, Claudio Freire <klaussfreire@gmail.com> wrote:
> > Maybe this is looking at the problem from the wrong direction.
> >
> > Why can't the page be added to the FSM immediately and the check be
> > done at runtime when looking for a reusable page?
> >
> > Index FSMs currently store only 0 or 255, couldn't they store 128 for
> > half-recyclable pages and make the caller re-check reusability before
> > using it?
> 
> No, because it's impossible for them to know whether or not the page
> that their index scan just landed on recycled just a second ago, or
> was like this since before their xact began/snapshot was acquired.
> 
> For your reference, this RecentGlobalXmin interlock stuff is what
> Lanin & Shasha call "The Drain Technique" within "2.5 Freeing Empty
> Nodes". Seems pretty hard to do it any other way.

Anyway(:p) the attached first patch is a PoC for the
cleanup-state-in-stats method works only for btree. Some
LOG-level debugging messages are put in the patch to show how it
works.

The following steps makes a not-recyclable page but I'm not sure
it is general enough, and I couldn't generate half-dead pages.
The pg_sleep() in the following steps is inserted in order to see
the updated values in stats.


DROP TABLE IF EXISTS t1;
CREATE TABLE t1 (a int);
CREATE INDEX ON t1 (a);
INSERT INTO t1 (SELECT a FROM generate_series(0, 800000) a);
DELETE FROM t1 WHERE a > 416700 AND a < 417250;
VACUUM t1;
DELETE FROM t1;
VACUUM t1;  -- 1 (or wait for autovacuum)
select pg_sleep(1);
VACUUM t1;  -- 2 (autovacuum doesn't work)
select pg_sleep(1);
VACUUM t1;  -- 3 (ditto)


The following logs are emited while the three VACUUMs are issued.

# VACUUM t1;  -- 1 (or wait for autovacuum)
 LOG:  btvacuumscan(t1_a_idx) result: deleted = 2185, notrecyclable = 1, hafldead = 0, no_cleanup_needed = false
 LOG:  Vacuum cleanup of index t1_a_idx is NOT skipped
 LOG:  btvacuumcleanup on index t1_a_idx is skipped since bulkdelete has run just before.
# VACUUM t1;  -- 2
 LOG:  Vacuum cleanup of index t1_a_idx is NOT skipped
 LOG:  btvacuumscan(t1_a_idx) result: deleted = 2192, notrecyclable = 0, hafldead = 0, no_cleanup_needed = true
# VACUUM t1;  -- 3
 LOG:  Vacuum cleanup of index t1_a_idx is skipped


VACUUM #1 leaves a unrecyclable page and requests the next cleanup.
VACUUM #2 leaves no unrecyclable page and inhibits the next cleanup.
VACUUM #3 (and ever after) no vacuum cleanup executed.

# I suppose it is a known issue that the cleanup cycles are not
# executed automatically unless new dead tuples are generated.

- Getting stats takes a very long time to fail during
  initdb. Since I couldn't find the right way to cope with this,
  I added a tentative function pgstat_live(), which checks that
  the backend has a valid stats socket.

- The patch calls pg_stat_get_vac_cleanup_needed using
  DirectFunctionCall. It might be better be wrapped.


As a byproduct, this enables us to run extra autovacuum rounds fo
r index cleanup. With the second attached, autovacuum works as
follows.

DROP TABLE IF EXISTS t1;
CREATE TABLE t1 (a int);
CREATE INDEX ON t1 (a);
INSERT INTO t1 (SELECT a FROM generate_series(0, 800000) a);
DELETE FROM t1 WHERE a > 416700 AND a < 417250;
(autovacuum on t1 runs)
> LOG:  btvacuumscan(t1_a_idx) result: deleted = 0, notrecyclable = 0, hafldead = 0, no_cleanup_needed = true
> LOG:  Vacuum cleanup of index t1_a_idx is skipped
> LOG:  automatic vacuum of table "postgres.public.t1": index scans: 1
DELETE FROM t1;
(autovacuum on t1 runs)
> LOG:  btvacuumscan(t1_a_idx) result: deleted = 2185, notrecyclable = 1, hafldead = 0, no_cleanup_needed = false
> LOG:  Vacuum cleanup of index t1_a_idx is NOT skipped
> LOG:  btvacuumcleanup on index t1_a_idx is skipped since bulkdelete has run just before.
> LOG:  automatic vacuum of table "postgres.public.t1": index scans: 1
(cleanup vacuum runs for t1 in the next autovac timing)
> LOG:  Vacuum cleanup of index t1_a_idx is NOT skipped
> LOG:  btvacuumscan(t1_a_idx) result: deleted = 2192, notrecyclable = 0, hafldead = 0, no_cleanup_needed = true
> LOG:  automatic vacuum of table "postgres.public.t1": index scans: 0


Any suggestions are welcome.

regards,

-- 
Kyotaro Horiguchi
NTT Open Source Software Center