postgresql/src/backend/optimizer/plan
Alvaro Herrera 0ac5ad5134 Improve concurrency of foreign key locking
This patch introduces two additional lock modes for tuples: "SELECT FOR
KEY SHARE" and "SELECT FOR NO KEY UPDATE".  These don't block each
other, in contrast with already existing "SELECT FOR SHARE" and "SELECT
FOR UPDATE".  UPDATE commands that do not modify the values stored in
the columns that are part of the key of the tuple now grab a SELECT FOR
NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently
with tuple locks of the FOR KEY SHARE variety.

Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this
means the concurrency improvement applies to them, which is the whole
point of this patch.

The added tuple lock semantics require some rejiggering of the multixact
module, so that the locking level that each transaction is holding can
be stored alongside its Xid.  Also, multixacts now need to persist
across server restarts and crashes, because they can now represent not
only tuple locks, but also tuple updates.  This means we need more
careful tracking of lifetime of pg_multixact SLRU files; since they now
persist longer, we require more infrastructure to figure out when they
can be removed.  pg_upgrade also needs to be careful to copy
pg_multixact files over from the old server to the new, or at least part
of multixact.c state, depending on the versions of the old and new
servers.

Tuple time qualification rules (HeapTupleSatisfies routines) need to be
careful not to consider tuples with the "is multi" infomask bit set as
being only locked; they might need to look up MultiXact values (i.e.
possibly do pg_multixact I/O) to find out the Xid that updated a tuple,
whereas they previously were assured to only use information readily
available from the tuple header.  This is considered acceptable, because
the extra I/O would involve cases that would previously cause some
commands to block waiting for concurrent transactions to finish.

Another important change is the fact that locking tuples that have
previously been updated causes the future versions to be marked as
locked, too; this is essential for correctness of foreign key checks.
This causes additional WAL-logging, also (there was previously a single
WAL record for a locked tuple; now there are as many as updated copies
of the tuple there exist.)

With all this in place, contention related to tuples being checked by
foreign key rules should be much reduced.

As a bonus, the old behavior that a subtransaction grabbing a stronger
tuple lock than the parent (sub)transaction held on a given tuple and
later aborting caused the weaker lock to be lost, has been fixed.

Many new spec files were added for isolation tester framework, to ensure
overall behavior is sane.  There's probably room for several more tests.

There were several reviewers of this patch; in particular, Noah Misch
and Andres Freund spent considerable time in it.  Original idea for the
patch came from Simon Riggs, after a problem report by Joel Jacobson.
Most code is from me, with contributions from Marti Raudsepp, Alexander
Shulgin, Noah Misch and Andres Freund.

This patch was discussed in several pgsql-hackers threads; the most
important start at the following message-ids:
	AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com
	1290721684-sup-3951@alvh.no-ip.org
	1294953201-sup-2099@alvh.no-ip.org
	1320343602-sup-2290@alvh.no-ip.org
	1339690386-sup-8927@alvh.no-ip.org
	4FE5FF020200002500048A3D@gw.wicourts.gov
	4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 12:04:59 -03:00
..
analyzejoins.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00
createplan.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00
initsplan.c Improve concurrency of foreign key locking 2013-01-23 12:04:59 -03:00
Makefile Remove cvs keywords from all files. 2010-09-20 22:08:53 +02:00
planagg.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00
planmain.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00
planner.c Improve concurrency of foreign key locking 2013-01-23 12:04:59 -03:00
README Remove useless whitespace at end of lines 2010-11-23 22:34:55 +02:00
setrefs.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00
subselect.c Update copyrights for 2013 2013-01-01 17:15:01 -05:00

src/backend/optimizer/plan/README

Subselects
==========

Vadim B. Mikheev


From owner-pgsql-hackers@hub.org Fri Feb 13 09:01:19 1998
Received: from renoir.op.net (root@renoir.op.net [209.152.193.4])
	by candle.pha.pa.us (8.8.5/8.8.5) with ESMTP id JAA11576
	for <maillist@candle.pha.pa.us>; Fri, 13 Feb 1998 09:01:17 -0500 (EST)
Received: from hub.org (hub.org [209.47.148.200]) by renoir.op.net (o1/$Revision: 1.14 $) with ESMTP id IAA09761 for <maillist@candle.pha.pa.us>; Fri, 13 Feb 1998 08:41:22 -0500 (EST)
Received: from localhost (majordom@localhost) by hub.org (8.8.8/8.7.5) with SMTP id IAA08135; Fri, 13 Feb 1998 08:40:17 -0500 (EST)
Received: by hub.org (TLB v0.10a (1.23 tibbs 1997/01/09 00:29:32)); Fri, 13 Feb 1998 08:38:42 -0500 (EST)
Received: (from majordom@localhost) by hub.org (8.8.8/8.7.5) id IAA06646 for pgsql-hackers-outgoing; Fri, 13 Feb 1998 08:38:35 -0500 (EST)
Received: from dune.krasnet.ru (dune.krasnet.ru [193.125.44.86]) by hub.org (8.8.8/8.7.5) with ESMTP id IAA04568 for <hackers@postgreSQL.org>; Fri, 13 Feb 1998 08:37:16 -0500 (EST)
Received: from sable.krasnoyarsk.su (dune.krasnet.ru [193.125.44.86])
	by dune.krasnet.ru (8.8.7/8.8.7) with ESMTP id UAA13717
	for <hackers@postgreSQL.org>; Fri, 13 Feb 1998 20:51:03 +0700 (KRS)
	(envelope-from vadim@sable.krasnoyarsk.su)
Message-ID: <34E44FBA.D64E7997@sable.krasnoyarsk.su>
Date: Fri, 13 Feb 1998 20:50:50 +0700
From: "Vadim B. Mikheev" <vadim@sable.krasnoyarsk.su>
Organization: ITTS (Krasnoyarsk)
X-Mailer: Mozilla 4.04 [en] (X11; I; FreeBSD 2.2.5-RELEASE i386)
MIME-Version: 1.0
To: PostgreSQL Developers List <hackers@postgreSQL.org>
Subject: [HACKERS] Subselects are in CVS...
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-pgsql-hackers@hub.org
Precedence: bulk
Status: OR

This is some implementation notes and opened issues...

First, implementation uses new type of parameters - PARAM_EXEC - to deal
with correlation Vars. When query_planner() is called, it first tries to
replace all upper queries Var referenced in current query with Param of
this type. Some global variables are used to keep mapping of Vars to
Params and Params to Vars.

After this, all current query' SubLinks are processed: for each SubLink
found in query' qual union_planner() (old planner() function) will be
called to plan corresponding subselect (union_planner() calls
query_planner() for "simple" query and supports UNIONs). After subselect
are planned, optimizer knows about is this correlated, un-correlated or
_undirect_ correlated (references some grand-parent Vars but no parent
ones: uncorrelated from the parent' point of view) query.

For uncorrelated and undirect correlated subqueries of EXPRession or
EXISTS type SubLinks will be replaced with "normal" clauses from
SubLink->Oper list (I changed this list to be list of EXPR nodes,
not just Oper ones). Right sides of these nodes are replaced with
PARAM_EXEC parameters. This is second use of new parameter type.
At run-time these parameters get value from result of subquery
evaluation (i.e. - from target list of subquery). Execution plan of
subquery itself becomes init plan of parent query. InitPlan knows
what parameters are to get values from subquery' results and will be
executed "on-demand" (for query select * from table where x > 0 and
y > (select max(a) from table_a) subquery will not be executed at all
if there are no tuples with x > 0 _and_ y is not used in index scan).

SubLinks for subqueries of all other types are transformed into
new type of Expr node - SUBPLAN_EXPR. Expr->args are just correlation
variables from _parent_ query. Expr->oper is new SubPlan node.

This node is used for InitPlan too. It keeps subquery range table,
indices of Params which are to get value from _parent_ query Vars
(i.e. - from Expr->args), indices of Params into which subquery'
results are to be substituted (this is for InitPlans), SubLink
and subquery' execution plan.

Plan node was changed to know about dependencies on Params from
parent queries and InitPlans, to keep list of changed Params
(from the above) and so be re-scanned if this list is not NULL.
Also, added list of InitPlans (actually, all of them for current
query are in topmost plan node now) and other SubPlans (from
plan->qual) - to initialize them and let them know about changed
Params (from the list of their "interests").

After all SubLinks are processed, query_planner() calls qual'
canonificator and does "normal" work. By using Params optimizer
is mostly unchanged.

Well, Executor. To get subplans re-evaluated without ExecutorStart()
and ExecutorEnd() (without opening and closing relations and indices
and without many palloc() and pfree() - this is what SQL-funcs does
on each call) ExecReScan() now supports most of Plan types...

Explanation of EXPLAIN.

vac=> explain select * from tmp where x >= (select max(x2) from test2
where y2 = y and exists (select * from tempx where tx = x));
NOTICE:  QUERY PLAN:

Seq Scan on tmp  (cost=40.03 size=101 width=8)
  SubPlan
  ^^^^^^^ subquery is in Seq Scan' qual, its plan is below
    ->  Aggregate  (cost=2.05 size=0 width=0)
          InitPlan
          ^^^^^^^^ EXISTS subsubquery is InitPlan of subquery
            ->  Seq Scan on tempx  (cost=4.33 size=1 width=4)
          ->  Result  (cost=2.05 size=0 width=0)
              ^^^^^^ EXISTS subsubquery was transformed into Param
                     and so we have Result node here
                ->  Index Scan on test2  (cost=2.05 size=1 width=4)


Opened issues.

1. No read permissions checking (easy, just not done yet).
2. readfuncs.c can't read subplan-s (easy, not critical, because of
   we currently nowhere use ascii representation of execution plans).
3. ExecReScan() doesn't support all plan types. At least support for
   MergeJoin has to be implemented.
4. Memory leaks in ExecReScan().
5. I need in advice: if subquery introduced with NOT IN doesn't return
   any tuples then qualification is failed, yes ?
6. Regression tests !!!!!!!!!!!!!!!!!!!!
   (Could we use data/queries from MySQL' crash.me ?
    Copyright-ed ? Could they give us rights ?)
7. Performance.
   - Should be good when subquery is transformed into InitPlan.
   - Something should be done for uncorrelated subqueries introduced
     with ANY/ALL - keep thinking. Currently, subplan will be re-scanned
     for each parent tuple - very slow...

Results of some test. TMP is table with x,y (int4-s), x in 0-9,
y = 100 - x, 1000 tuples (10 duplicates of each tuple). TEST2 is table
with x2, y2 (int4-s), x2 in 1-99, y2 = 100 -x2, 10000 tuples (100 dups).

   Trying

select * from tmp where x >= (select max(x2) from test2 where y2 = y);

   and

begin;
select y as ty, max(x2) as mx into table tsub from test2, tmp
where y2 = y group by ty;
vacuum tsub;
select x, y from tmp, tsub where x >= mx and y = ty;
drop table tsub;
end;

   Without index on test2(y2):

SubSelect         -> 320 sec
Using temp table  -> 32 sec

   Having index

SubSelect         -> 17 sec (2M of memory)
Using temp table  -> 32 sec (12M of memory: -S 8192)

Vadim