Index condition in a Nested Loop

From: Mark Hills <mark(at)pogo(dot)org(dot)uk>
To: pgsql-performance(at)postgresql(dot)org
Subject: Index condition in a Nested Loop
Date: 2012-02-26 14:16:47
Message-ID: alpine.LNX.2.01.1202261358070.31287@stax.localdomain
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

I found in a production system that mostly the performance is being
crippled by the handling of one particular case.

I hope I've reduced this to a minimal example, below.

Summary: when more than one key is given (or the data is sourced from a
table) the planner is forced to a join of the whole data set.

The query aggregates a small amount of the data in a large data set. It
looks like lookups via a Nested Loop would be considerably quicker. I did
this explicitly in the UNION query.

What is that prevents the index condition from being used in earlier parts
of the query? Only where a single condition is present is it be used below
the final join.

Is it that a Nested Loop cannot make several independent lookups via an
index?

Is my input preventing the planner doing this, or would it need to be
smarter about something?

It seems interesting that it is able to do this successfully in the third
plan: "As above but without the join to the job table".

Thanks

--
Mark

--
-- Create schema: job -E task -E resource
--

CREATE TABLE job (
jid integer PRIMARY KEY
);

CREATE TABLE task (
jid integer REFERENCES job (jid),
tid integer PRIMARY KEY
);

CREATE TABLE resource (
tid integer REFERENCES task (tid),
name text
);

CREATE INDEX idx_task ON task (jid, tid);
CREATE INDEX idx_resource_tid ON resource (tid, name);
CREATE INDEX idx_resource_name ON resource (name, tid);

--
-- Populate with data:
-- 10000 jobs,
-- 1000 tasks per job,
-- 0-4 resources per task
--

CREATE OR REPLACE FUNCTION
populate()
RETURNS VOID
AS $$
DECLARE
t integer;
BEGIN
FOR t IN 0..10000 LOOP
INSERT INTO job VALUES (t);
END LOOP;

FOR t IN 0..10000000 LOOP
INSERT INTO task VALUES (random() * 10000, t);

IF random() > 0.1 THEN
INSERT INTO resource VALUES (t, 'wallace');
INSERT INTO resource VALUES (t, 'gromit');
END IF;

IF random() > 0.9 THEN
INSERT INTO resource VALUES (t, 'shaun');
END IF;

IF random() > 0.6 THEN
INSERT INTO resource VALUES (t, 'wendolene');
END IF;
END LOOP;
END
$$ LANGUAGE plpgsql;

SELECT populate();
VACUUM ANALYZE;

-- Define some simple aggregation with a left join

CREATE VIEW middle AS
SELECT task.jid,
task.tid,
COUNT(resource.name) AS nresource
FROM task
LEFT JOIN resource ON task.tid = resource.tid
GROUP BY task.jid,
task.tid;

-- Aggregate again for a single key: fast
-- "Nested Loop" is used

SELECT job.jid,
sum(nresource)
FROM job
INNER JOIN middle ON job.jid = middle.jid
WHERE job.jid IN (1234)
GROUP BY job.jid;

-- GroupAggregate (cost=0.00..35026.04 rows=1 width=12)
-- -> Nested Loop (cost=0.00..35021.13 rows=980 width=12)
-- -> Index Only Scan using job_pkey on job (cost=0.00..4.28 rows=1 width=4)
-- Index Cond: (jid = 1234)
-- -> GroupAggregate (cost=0.00..34997.25 rows=980 width=15)
-- -> Nested Loop Left Join (cost=0.00..34970.55 rows=2254 width=15)
-- -> Index Only Scan using idx_task on task (cost=0.00..55.98 rows=980 width=8)
-- Index Cond: (jid = 1234)
-- -> Index Only Scan using idx_resource_tid on resource (cost=0.00..35.54 rows=7 width=11)
-- Index Cond: (tid = task.tid)
-- (10 rows)

-- As above, but with two keys: slow
-- "Merge Join" is attempted; this is the 'bad' case

EXPLAIN
SELECT job.jid,
sum(nresource)
FROM job
INNER JOIN middle ON job.jid = middle.jid
WHERE job.jid IN (1234, 5678)
GROUP BY job.jid;

-- GroupAggregate (cost=5636130.95..6091189.12 rows=2 width=12)
-- -> Merge Join (cost=5636130.95..6091179.10 rows=2000 width=12)
-- Merge Cond: (task.jid = job.jid)
-- -> GroupAggregate (cost=5636130.95..5966140.73 rows=9999986 width=15)
-- -> Sort (cost=5636130.95..5693633.43 rows=23000992 width=15)
-- Sort Key: task.jid, task.tid
-- -> Merge Left Join (cost=0.00..1251322.49 rows=23000992 width=15)
-- Merge Cond: (task.tid = resource.tid)
-- -> Index Scan using task_pkey on task (cost=0.00..281847.80 rows=9999986 width=8)
-- -> Index Only Scan using idx_resource_tid on resource (cost=0.00..656962.32 rows=23000992 width=11)
-- -> Materialize (cost=0.00..8.55 rows=2 width=4)
-- -> Index Only Scan using job_pkey on job (cost=0.00..8.54 rows=2 width=4)
-- Index Cond: (jid = ANY ('{1234,5678}'::integer[]))
-- (13 rows)

-- As above but without the join to the job table: fast

SELECT jid,
sum(nresource)
FROM middle
WHERE jid IN (1234, 5678)
GROUP BY jid;

-- GroupAggregate (cost=0.00..69995.03 rows=200 width=12)
-- -> GroupAggregate (cost=0.00..69963.62 rows=1961 width=15)
-- -> Nested Loop Left Join (cost=0.00..69910.18 rows=4511 width=15)
-- -> Index Only Scan using idx_task on task (cost=0.00..93.39 rows=1961 width=8)
-- Index Cond: (jid = ANY ('{1234,5678}'::integer[]))
-- -> Index Only Scan using idx_resource_tid on resource (cost=0.00..35.52 rows=7 width=11)
-- Index Cond: (tid = task.tid)
-- (7 rows)

-- Kludge to lookup two keys: fast (cost 70052)

SELECT job.jid,
sum(nresource)
FROM job
INNER JOIN middle ON job.jid = middle.jid
WHERE job.jid IN (1234)
GROUP BY job.jid
UNION
SELECT job.jid,
sum(nresource)
FROM job
INNER JOIN middle ON job.jid = middle.jid
WHERE job.jid IN (5678)
GROUP BY job.jid;

--
-- Repeat with job keys from a table instead of 'IN' clause.
--

CREATE TABLE one_job (
jid integer PRIMARY KEY
);

CREATE TABLE two_jobs (
jid integer PRIMARY KEY
);

INSERT INTO one_job VALUES (1234);
INSERT INTO two_jobs VALUES (1234), (5678);

ANALYZE one_job;
ANALYZE two_jobs;

-- Joining against one row: slow (cost 5636131.97..6092141.59)
-- "Merge Join" is attempted

EXPLAIN
SELECT job.jid,
sum(nresource)
FROM one_job job
INNER JOIN middle ON job.jid = middle.jid
GROUP BY job.jid;

-- Joining against two rows: slow (cost 5636131.98..6093141.60)
-- "Merge Join" is attempted

EXPLAIN
SELECT job.jid,
sum(nresource)
FROM two_jobs job
INNER JOIN middle ON job.jid = middle.jid
GROUP BY job.jid;

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message Andy Colson 2012-02-26 15:20:54 Re: PG as in-memory db? How to warm up and re-populate buffers? How to read in all tuples into memory?
Previous Message Stephen Frost 2012-02-26 13:55:11 Re: PG as in-memory db? How to warm up and re-populate buffers? How to read in all tuples into memory?