Re: BUG #6785: Memory Leak in plpgsql - Mailing list pgsql-bugs

Hi,


2012/7/30 Andres Freund <andres@2ndquadrant.com>

> Hi,
>
> On Monday, July 30, 2012 03:15:37 PM andervalbh@gmail.com wrote:
> >   we recently had a memory exhaustion in the PostgreSQL server of the
> > company, after a scan found a likely memory leak when using a plpgsql
> > function.
> > The problem occurred on an IBM x3400 server with 12G, CentOS 5.5 and
> > PostgreSQL 9.1.4. The leak occurs when a variable declared type of a
> table
> > column and when the
> > column or the table ceases to exist.
> > Follow the steps for the simulation:
> >
> > create table tbl_test
> > (cod integer);
> >
> > CREATE OR REPLACE FUNCTION citgis.fct_test()
> > RETURNS void AS
> > $body$
> > DECLARE
> >   v_cod tbl_test.cod%type;
> > BEGIN
> >   return;
> > END;
> > $body$
> > LANGUAGE 'plpgsql';
> >
> > drop table tbl_test;
> >
> >
> > test=# select pg_backend_pid();
> >  pg_backend_pid
> > ----------------
> >            6465
> >
> > Initial memory
> >
> ---------------------------------------------------------------------------
> > --------------------------------------------- PID USER      PR  NI  VIRT
> > SWAP  RES CODE DATA  SHR S P %CPU %MEM    TIME+ COMMAND
> > 6465 postgres  15   0 1194m 1.2g  14m 4684  12m 3748 S 1  0.0  0.1
> > 0:00.08 postgres: postgres test [local] idle
> > 6465 postgres  15   0 1203m 1.2g  23m 4684  21m 3748 S 1  0.0  0.2
> > 0:00.17 postgres: postgres test [local] idle
> > 6465 postgres  15   0 1212m 1.2g  32m 4684  30m 3748 S 1  3.0  0.3
> > 0:00.26 postgres: postgres test [local] idle
> > 6465 postgres  16   0 1221m 1.2g  41m 4684  39m 3756 S 1  0.0  0.3
> > 0:00.36 postgres: postgres test [local] idle
> >
> ---------------------------------------------------------------------------
> > ---------------------------------------------
> I don't think youve found a memory leak here. I tested several thousand
> iterations of this and the memory usage tops out a 93MB. What you see is
> probably some memory fragmentation and that some copy-on-write page
> mappings
> are only gradually brought into place.
>
> Greetings,
>
> Andres
> --
>  Andres Freund                     http://www.2ndQuadrant.com/
>  PostgreSQL Development, 24x7 Support, Training & Services
>

I understand, but the memory should not be returned after the execution of
the function?
Below is the result of running with more than 55,000 calls.

PID USER      PR  NI  VIRT SWAP  RES CODE DATA  SHR S P %CPU %MEM    TIME+
 COMMAND
618 postgres  15   0 1687m 1.2g 507m 4684 505m 3796 S 1  0.0  4.2   0:04.90
postgres: postgres test [local] idle

Greetings,

  Anderson