PostgreSQL memory leak

README.md

This was performed on Ubuntu 14.04 with a fresh install of PostgreSQL 9.3.14 directly from the official Postgres apt repo. It only works if the query planner chooses the attached plan (with HashAggregate).

Create and populate a new database:

psql -U postgres -c "create database test;"
psql -U postgres test < populate.sql

Open a connection to the database and execute query.sql. The Postgres worker memory usage will rise approximately 50MB and won't be released until the connection is closed.

populate.sql

CREATE TABLE venue(
	id INTEGER PRIMARY KEY
);

CREATE TABLE guest(
	id SERIAL PRIMARY KEY,
	venue_id INTEGER NOT NULL REFERENCES venue(id)
);

CREATE TABLE reservation(
	id SERIAL PRIMARY KEY,
	guest_id INTEGER NOT NULL REFERENCES guest(id),
	venue_id INTEGER NOT NULL REFERENCES venue(id)
);

CREATE FUNCTION populate() RETURNS VOID AS $$
DECLARE
	i INTEGER;
	j INTEGER;
	k INTEGER;
	guest_id INTEGER;
BEGIN
	FOR i IN 1..10 LOOP
		INSERT INTO venue VALUES (i);

		FOR j IN 1..1000 LOOP
			INSERT INTO guest(venue_id) VALUES (i) RETURNING id INTO guest_id;

			FOR k IN 1..10 LOOP
				INSERT INTO reservation(guest_id, venue_id) VALUES (guest_id, i);
			END LOOP;
		END LOOP;
	END LOOP;
END;
$$ LANGUAGE plpgsql;
SELECT populate();

query-plan.txt

                                                         QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------
 HashAggregate  (cost=2364.30..2490.68 rows=10110 width=8) (actual time=29.270..33.212 rows=10000 loops=1)
   ->  Hash Join  (cost=270.00..2313.75 rows=10110 width=8) (actual time=4.263..18.959 rows=10000 loops=1)
         Hash Cond: (reservation.guest_id = guest.id)
         ->  Seq Scan on reservation  (cost=0.00..1791.00 rows=10110 width=8) (actual time=0.007..10.111 rows=10000 loops=1)
               Filter: (venue_id = 1)
               Rows Removed by Filter: 90000
         ->  Hash  (cost=145.00..145.00 rows=10000 width=4) (actual time=4.250..4.250 rows=10000 loops=1)
               Buckets: 1024  Batches: 1  Memory Usage: 352kB
               ->  Seq Scan on guest  (cost=0.00..145.00 rows=10000 width=4) (actual time=0.004..2.063 rows=10000 loops=1)
 Total runtime: 35.944 ms
(10 rows)

query.sql

SELECT reservation.id,
        array_agg(guest.id) AS guest_ids
FROM reservation

JOIN guest ON guest.id = reservation.guest_id

WHERE reservation.venue_id = 1
GROUP BY reservation.id;

setup.sh

sudo apt-get update && sudo apt-get upgrade -y
sudo bash -c 'echo "deb http://apt.postgresql.org/pub/repos/apt/ trusty-pgdg main" > /etc/apt/sources.list.d/pgdg.list'
wget --quiet -O - https://www.postgresql.org/media/keys/ACCC4CF8.asc | \
  sudo apt-key add -
sudo apt-get update
sudo apt-get install -y postgresql-9.3 postgresql-9.3-dbg
sudo -u postgres psql -c "create database test;"
curl https://gist.githubusercontent.com/luhn/2b35a9b31255e3a6a2e6a06d1213dfc9/raw/d7a4fb433c3bb8f60f0e12e502adef39e57d4369/populate.sql | sudo -u postgres psql test
sudo vim /etc/postgresql/9.3/main/postgresql.conf  # Adjust work_mem to 128MB

This looks very interesting to me. It seems the memory leak was introduced and was removed without a user report, which means the developers were unaware of the issue until you reported it.

That seems to be jumping to conclusions. Maybe there was a user report, but you just didn't see it? The message that was quoted above mentions that a fix was done in
https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=b419865a814abbca12bdd6eef6a3d5ed67f432e1
(which was part of Postgres 9.5.0 and onwards) and in that commit message Jeff cites this 2013 report: https://postgr.es/m/3839201.Nfa2RvcheX@techfox.foxi