lkml.org 
[lkml]   [2011]   [Apr]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [BUG] fatal hang untarring 90GB file, possibly writeback related.
On Thu, Apr 28, 2011 at 02:59:27PM -0500, James Bottomley wrote:
> On Thu, 2011-04-28 at 20:21 +0100, Mel Gorman wrote:
> > On Thu, Apr 28, 2011 at 01:30:36PM -0500, James Bottomley wrote:
> > > > Way hey, cgroups are also in the mix. How jolly.
> > > >
> > > > Is systemd a common element of the machines hitting this bug by any
> > > > chance?
> > >
> > > Well, yes, the bug report is against FC15, which needs cgroups for
> > > systemd.
> > >
> >
> > Ok although we do not have direct evidence that it's the problem yet. A
> > broken shrinker could just mean we are also trying to aggressively
> > reclaim in cgroups.
> >
> > > > The remaining traces seem to be follow-on damage related to the three
> > > > issues of "shrinkers are bust in some manner" causing "we are not
> > > > getting over the min watermark" and as a side-show "we are spending lots
> > > > of time doing something unspecified but unhelpful in cgroups".
> > >
> > > Heh, well find a way for me to verify this: I can't turn off cgroups
> > > because systemd then won't work and the machine won't boot ...
> > >
> >
> > Same testcase, same kernel but a distro that is not using systemd to
> > verify if cgroups are the problem. Not ideal I know. When I'm back
> > online Tuesday, I'll try reproducing this on a !Fedora distribution. In
> > the meantime, the following untested hatchet job might spit out
> > which shrinker we are getting stuck in. It is also breaking out of
> > the shrink_slab loop so it'd even be interesting to see if the bug
> > is mitigated in any way.
>
> Actually, talking to Chris, I think I can get the system up using
> init=/bin/bash without systemd, so I can try the no cgroup config.
>
> > diff --git a/mm/vmscan.c b/mm/vmscan.c
> > index c74a501..ed99104 100644
>
> In the mean time, this patch produces:
>
> (that's nothing ... apparently the trace doesn't activate when kswapd
> goes mad).
>

Or is looping there for shorter than we expect. HZ/10?

--
Mel Gorman
SUSE Labs


\
 
 \ /
  Last update: 2011-04-28 22:29    [W:0.127 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site