I'm running 2.2.17 vanilla on a UP x86 box, and getting occasionally a
couple of 'VM: do_try_to_free_pages failed' messages.
The system appears to be running perfectly. It's almost out of real RAM,
but has about 100M swap unused.
I can't figure out how this happens. Specifically, how come the call to
swap_out in do_try_to_free_pages didn't swap something out, return true,
and avoid that message being printed?
kswapd is not acting up in any way; the system doesn't appear to be OOM.
If this isn't a bug then can we remove this printk'd message?
If it does seem to be a bug and someone'll give me a crash course on this
area of the VM I'll investigate further. I notice do_try_to_free_pages can
be called either from kswapd, or under what look like memory-pressure
conditions elsewhere.
I'm not on lkml, so please cc me on any replies.
--matt
From: "Matthew Sanderson" <[email protected]>
> I'm running 2.2.17 vanilla on a UP x86 box, and getting occasionally a
> couple of 'VM: do_try_to_free_pages failed' messages.
> The system appears to be running perfectly. It's almost out of real RAM,
> but has about 100M swap unused.
>
> I can't figure out how this happens. Specifically, how come the call to
> swap_out in do_try_to_free_pages didn't swap something out, return true,
> and avoid that message being printed?
> kswapd is not acting up in any way; the system doesn't appear to be OOM.
>
> If this isn't a bug then can we remove this printk'd message?
> If it does seem to be a bug and someone'll give me a crash course on this
> area of the VM I'll investigate further. I notice do_try_to_free_pages can
> be called either from kswapd, or under what look like memory-pressure
> conditions elsewhere.
>
> I'm not on lkml, so please cc me on any replies.
>
> --matt
>
I have this problem with one of my servers with kernel 2.2.16, and then
postgresql freaked out saying something like "AllocSet: memory exhausted".
I changed to 2.2.17, and the messages stopped turning up in the logs, but
pgsql still failes sometimes.
I have to restart the daemon completely to get it working again.
I dunno if there's something wrong with pgsql's memory/shared mem
management, or if it's the kernels fault.
(On 2.0.38 pgsql(different version tho) + box has a 700 day uptime)
Example log (2.2.16):
Oct 24 00:01:34 gimme kernel: VM: do_try_to_free_pages failed for
postmaster...
Oct 24 00:01:34 gimme kernel: VM: do_try_to_free_pages failed for klogd...
Oct 24 00:01:34 gimme kernel: VM: do_try_to_free_pages failed for caspd...
Oct 24 00:07:39 gimme kernel: VM: do_try_to_free_pages failed for
postmaster...
Magnus Naeslund
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Programmer/Networker [|] Magnus Naeslund
PGP Key: http://www.genline.nu/mag_pgp.txt
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> Oct 24 00:07:39 gimme kernel: VM: do_try_to_free_pages failed for
> postmaster...
2.2.18pre19 should fix this problem if andrea's patch is inside.
if not, you have to patch pre18 with VM-global-2.2.18pre18-7.bz2
if you are from europe you can downlaod it from:
ftp://ftp.ovh.net/pub/linux/kernel/people/andrea/patches/v2.2/2.2.18pre18/VM-global-2.2.18pre18-7.bz2
Octave
From: "octave klaba" <[email protected]>
> > Oct 24 00:07:39 gimme kernel: VM: do_try_to_free_pages failed for
> > postmaster...
>
> 2.2.18pre19 should fix this problem if andrea's patch is inside.
> if not, you have to patch pre18 with VM-global-2.2.18pre18-7.bz2
> if you are from europe you can downlaod it from:
>
ftp://ftp.ovh.net/pub/linux/kernel/people/andrea/patches/v2.2/2.2.18pre18/VM
-global-2.2.18pre18-7.bz2
>
> Octave
>
I can confirm that pre20 + arcangelis VM-global seems to apply fine.
No troubles here, compiling it now...
Magnus
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Programmer/Networker [|] Magnus Naeslund
PGP Key: http://www.genline.nu/mag_pgp.txt
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-