From: Reindl Harald
Date: April 13 2012 2:35pm
Subject: Re: FYI: how to prevent mysql from oom-killer
List-Archive: http://lists.mysql.com/mysql/227192
Message-Id: <4F8839BC.5090805@thelounge.net>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha1;
 protocol="application/pgp-signature";
 boundary="------------enig217147D52D6FFFB10311EBBE"

--------------enig217147D52D6FFFB10311EBBE
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

hm - my servers generally have no swap or better said
only a 500 MB swap-file, they are all virtual machines
in a ESXi cluster

you really do not want a VM swapping
this leads in unusaeable performance

in my case the problem was another process eating up all memory
by user-mistake and my hint is to make sure in this case that
this process get killed and never mysqld

but i never ever had to reboot any linux guest to get it back to life

Am 13.04.2012 16:27, schrieb Johnny Withers:
> I had this same problem with the OOM for a while. Very frustrating to h=
ave to reboot a server to bring it back to life.
>=20
> I found out the OOM only ran when the swap file was about 99% full. The=
 servers I had this problem on had 16GB and
> 24GB of ram, but only 2GB of swap. I increased the swap on the 24GB ser=
vers to 48GB and on the 16GB servers to
> 32GB. The swap never fills to over 60% now and I haven't' had any OOM p=
roblems since and the systems run great.
>=20
> I've also set vm.swappiness=3D0 in /etc/sysctl.conf
>=20
> -JW
>=20
> On Fri, Apr 13, 2012 at 7:17 AM, Reindl Harald <h.reindl@stripped =
<mailto:h.reindl@stripped>> wrote:
>=20
>     the following may be useful for most server systems
>=20
>     OOM-killer acts if some process reclaims more and more
>     memory and the kernel randomly kills unimportant tasks
>     using hughe memory
>=20
>     in case of a running mysqld the classification "unimportant"
>     is nearly all time wrong and can cause hughe damage and work
>     in other words: you really never want killed a database server
>     randomly instead "dbmail-imapd" which can be restarted via
>     systemd without pain and may be the root-cause of OOM
>     _________________________________________
>=20
>     with one single command you can protect processes from get killed
>     i started to run this every 15 minutes to make sure it is also
>     active after restarts
>=20
>     i am considering include this in "mysqld.service" as
>     "ExecStartPost=3D-/usr/local/bin/mysql-no-oom.sh" in our
>     internal mysqld-packages and include also the script
>     _________________________________________
>=20
>     [root@mail:~]$ cat /etc/crontab | grep oom
>     0,15,30,45 * * * * root bash /usr/local/bin/mysql-no-oom.sh
>=20
>     [root@mail:~]$ cat /usr/local/bin/mysql-no-oom.sh
>     #!/bin/bash
>     pgrep -f "/usr/libexec/mysqld" | while read PID; do echo -1000 > /p=
roc/$PID/oom_score_adj; done
>=20
>=20
>=20
>=20
> --=20
> -----------------------------
> Johnny Withers
> 601.209.4985
> johnny@stripped <mailto:johnny@stripped>


--------------enig217147D52D6FFFB10311EBBE
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: OpenPGP digital signature
Content-Disposition: attachment; filename="signature.asc"

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk+IObwACgkQhmBjz394AnmtiACfdmpv9cu0mQJxvTW4qE4WLMt5
oN4An1mBxV5WrJHKioO6IpSf3TqFdu3t
=TSOB
-----END PGP SIGNATURE-----

--------------enig217147D52D6FFFB10311EBBE--