Re: 2.6.18-rc3-mm2 - OOM storm



On Thu, 10 Aug 2006 11:04:36 +0200
Laurent Riffard <laurent.riffard@xxxxxxx> wrote:

Le 06.08.2006 12:08, Andrew Morton a écrit :
ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.18-rc3/26.18-rc3-mm2/

Hello,

On my system, a cron runs every day to check the integrity of
installed RPMS, it runs "rpm -v" on each package, which computes
MD5 hash for each installed file and compares this result, the file
size and modification time with values stored in RPM database.

This is the workload. Since 2.6.18-rc3-mm2, this processus eats
all the memory and triggers OOM.

On my system, "free -t" output normally looks like this ("cached" value
is about half of RAM):
# free -t
total used free shared buffers cached
Mem: 515032 508512 6520 0 22992 256032
-/+ buffers/cache: 229488 285544
Swap: 1116428 324 1116104
Total: 1631460 508836 1122624

After the rpm database check, "free -t" says:
total used free shared buffers cached
Mem: 515032 507124 7908 0 8132 398296
-/+ buffers/cache: 100696 414336
Swap: 1116428 34896 1081532
Total: 1631460 542020 1089440

And the value of "cached" won't decrease.


Yes, I was just trying to reproduce this. No luck so far. Will try your
.config tomorrow.

It would be interesting to try disabling CONFIG_ADAPTIVE_READAHEAD -
perhaps that got broken.

Also, are you able to determine whether the problem is specific to `rpm
-V'? Are you able to make the leak trigger using other filesystem
workloads?

If it's specific to `rpm -V' then perhaps direct-io is somehow causing
pagecache leakage. That would be a bit odd.



btw, it's not necessary to go all the way to oom to work out if the
pagecache leak is happening. After booting, do

echo 3 > /proc/sys/vm/drop_pagecache

and record the `Cached' figure in /proc/meminfo. After running some test,
run `echo 3 > /proc/sys/vm/drop_pagecache' again and check
/proc/meminfo:Cached. If it dodn't do gown to a similarly low figure,
we're leaking pagecache.

btw2: please use /proc/meminfo output rather than free(1). Because free(1)
shows less info, and it does mysterious mangling of the info which it does
read in ways which confuse me.

Thanks.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



Relevant Pages

  • Re: 2.6.18-rc3-mm2 - OOM storm
    ... Laurent Riffard wrote: ... installed RPMS, it runs "rpm -v" on each package, which computes ... size and modification time with values stored in RPM database. ... pagecache leak is happening. ...
    (Linux-Kernel)
  • Re: 2.6.18-rc3-mm2 - OOM storm
    ... installed RPMS, it runs "rpm -v" on each package, which computes ... size and modification time with values stored in RPM database. ... total used free shared buffers cached ... Call Trace: ...
    (Linux-Kernel)
  • Re: apt losts my rpm installed list
    ... > Rpm -qa seems fine. ... But apt shows no installed rpms. ... > System is a Redhat 9.0, ...
    (comp.os.linux.misc)
  • Re: Issues with kernel 2.6.15 and other updates this morning
    ... Or based on a given window of time? ... rpm -qa --last | tac ... will show all currently installed RPMs in install date order ...
    (Fedora)
  • Re: RPM uninstall with Red Hat 9
    ... As far as command syntax goes there doesn't seems to be a problem. ... has its own database which keeps all the information regarding ... installed rpms. ... software through rpm command? ...
    (comp.os.linux.x)