Re: Leaking regexp_replace in 9.3.1 ? (was: [HACKERSUninterruptable regexp_replace in 9.3.1 ?) - Mailing list pgsql-bugs

From Sandro Santilli
Subject Re: Leaking regexp_replace in 9.3.1 ? (was: [HACKERSUninterruptable regexp_replace in 9.3.1 ?)
Date
Msg-id 20140319075359.GA4042@localhost
Whole thread Raw
In response to Re: Leaking regexp_replace in 9.3.1 ? (was: [HACKERSUninterruptable regexp_replace in 9.3.1 ?)  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: Leaking regexp_replace in 9.3.1 ? (was: [HACKERSUninterruptable regexp_replace in 9.3.1 ?)  (Sandro Santilli <strk@keybit.net>)
List pgsql-bugs
On Tue, Mar 18, 2014 at 07:41:32PM -0400, Tom Lane wrote:
> Sandro Santilli <strk@keybit.net> writes:
> > Tom: I saw you pushed a fix for the interruptability with this:
> > https://github.com/postgres/postgres/commit/f5f21315d25ffcbfe7c6a3fa6ffaad54d31bcde0
>
> > But we also noticed a memory leak in the regepx_replace call, did you notice
> > that in your tests ? Same regexp as reported. Do you need another testcase ?
>
> I see no memory growth while running your original example against HEAD.

This is what I'm running:

 SELECT regexp_matches('test', $REG$((?:[^'"$;]+|"[^"]*"|'(?:[^']*|'')*'|(\$[^$]*\$).*?\2)+)$REG$, 'g' );

And this is what I see as the process memory:

  PID TTY      STAT   TIME  MAJFL   TRS   DRS   RSS %MEM COMMAND
 5359 ?        Ss     0:00      0  5693 172446 2908  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 172718 4224  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 173250 4488  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 173670 5016  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 174070 5280  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 174494 5544  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 174762 5808  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 175690 6600  0.0 postgres: strk strk [local] idle
 5359 ?        Ss     0:00      0  5693 179006 9240  0.0 postgres: strk strk [local] idle

This is as of commit 63817f86b57fc3d29b57787bca9d786218b7ee25

Or, simpler, from this state:

  PID TTY      STAT   TIME  MAJFL   TRS   DRS   RSS %MEM COMMAND
 5359 ?        Ss     0:00      0  5693 1274962 869832  5.3 postgres: strk strk [local] idle

I run this:

 SELECT count(*) from (
   select regexp_matches( 'test', $REG$((?:[^'"$;]+|"[^"]*"|'(?:[^']*|'')*'|(\$[^$]*\$).*?\2)+)$REG$, 'g' )
   FROM generate_series(1, 10000)
 ) as f;

And get here:

  PID TTY      STAT   TIME  MAJFL   TRS   DRS   RSS %MEM COMMAND
 5359 ?        Ss     0:00      0  5693 1274962 869832  5.3 postgres: strk strk [local] idle
 5359 ?        Rs     0:00      0  5693 1555894 1090536  6.6 postgres: strk strk [local] SELECT
 5359 ?        Ss     0:00      0  5693 2269682 1650744 10.0 postgres: strk strk [local] idle

The memory is only released on session close.
This is with generate_series going up to 1e5:

  PID TTY      STAT   TIME  MAJFL   TRS   DRS   RSS %MEM COMMAND
 5359 ?        Ss     0:00      0  5693 2269682 1650744 10.0 postgres: strk strk [local] idle
 5359 ?        Rs     0:00      0  5693 2300726 1675560 10.2 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:01      0  5693 6438626 4923024 30.0 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:02      0  5693 10591778 8182368 50.0 postgres: strk strk [local] SELECT
 5359 ?        Ss     0:03      0  5693 12216566 9457488 57.8 postgres: strk strk [local] idle

Up to 1e6 the story ends with a crash, after:

  PID TTY      STAT   TIME  MAJFL   TRS   DRS   RSS %MEM COMMAND
 5359 ?        Ss     0:03      0  5693 12216566 9457488 57.8 postgres: strk strk [local] idle
 5359 ?        Rs     0:04      0  5693 15666674 12166392 74.3 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:05      0  5693 19403254 15099056 92.3 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:06      0  5693 20284546 15366412 93.9 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:07      1  5693 21665438 15333312 93.7 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:08      1  5693 23065426 15295472 93.5 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:09      1  5693 24468334 15256592 93.2 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:10      1  5693 25869382 15225544 93.0 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:11      1  5693 27280514 15210012 92.9 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:12      1  5693 28684350 15173964 92.7 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:12      1  5693 29683014 15100680 92.3 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:13      1  5693 30298130 15051580 92.0 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:13      1  5693 30979690 15088992 92.2 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:13      1  5693 31758466 15139792 92.5 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:14      1  5693 32406206 15111128 92.3 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:14      1  5693 33045722 15131676 92.5 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:14      1  5693 33479010 15135036 92.5 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:15      1  5693 33847174 15146748 92.6 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:15      1  5693 34186830 15131648 92.5 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:15      1  5693 34339878 15119460 92.4 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:15      1  5693 34502474 15139816 92.5 postgres: strk strk [local] SELECT
 5359 ?        Ds     0:15      1  5693 34625554 15143068 92.5 postgres: strk strk [local] SELECT
 5359 ?        Rs     0:15   1833     0     0     0  0.0 [postgres]

The connection to the server was lost. Attempting reset: Failed.
!>

The configure switches:

./configure \
        --prefix=/home/postgresql-9.3.4 \
        --with-libxml \
        --with-python \
        --enable-nls \
        --enable-cassert \
        --enable-debug \
        --with-ossp-uuid

--strk;

 ()  ASCII ribbon campaign  --  Keep it simple !
 /\  http://strk.keybit.net/rants/ascii_mails.txt

pgsql-bugs by date:

Previous
From: Michael Gerstein
Date:
Subject: Re: BUG #9478: Greenplum External Table definitions
Next
From: Jesse Denardo
Date:
Subject: Re: BUG #9616: Materialized view with indexes unable to load from pg_dump