[Top] [All Lists]

Re: xfs task blocked for more than 120 seconds

To: xfs@xxxxxxxxxxx
Subject: Re: xfs task blocked for more than 120 seconds
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Thu, 2 Feb 2012 11:26:08 +1100
In-reply-to: <20120131233034.GK10174@xxxxxxxxx>
References: <20120130002026.GG10174@xxxxxxxxx> <20120130010530.GI15102@dastard> <2504_1327964557_4F27218D_2504_92_2_20120130223527.GH10174@xxxxxxxxx> <20120131233034.GK10174@xxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Wed, Feb 01, 2012 at 01:30:34AM +0200, Sami Liedes wrote:
> On Tue, Jan 31, 2012 at 12:35:28AM +0200, Sami Liedes wrote:
> > The workloads do eventually complete. I tried the tar extraction again
> > but this time extracting the tar from a different disk and saw no such
> > warnings (and the time taken seems reasonable at 96 minutes).
> Extracting a 105 GB tar that resides on the same (target) XFS
> partition on an otherwise idle disk does trigger warnings, but only
> for kworker, not the tar process. The time taken to extract was a
> reasonable-sounding 140 minutes, so maybe these are just harmless
> warnings.

It is indicating that superblock IO is taking more than 120s to
complete. That's an indication that the IO subsystem is not
throttling the IO queue (and so queuing up minutes worth of IO) or
not processing the IO queue in a fair/timely manner. IO should never
take more than a couple of seconds to complete unless there is
something wrong, so the warnings are quite valid at the default
timeout of 120s....

I'd suggest reporting the problem to the dm-crypt folks and see if
they have any ideas on why IO completion is taking so long.

> Perhaps something should/could be done to silence them in
> that case?

You can turn the hangcheck timer off or adjust it's timeout yourself
via these proc files:

$ ls /proc/sys/kernel/ |grep hung


Dave Chinner

<Prev in Thread] Current Thread [Next in Thread>
  • Re: xfs task blocked for more than 120 seconds, Dave Chinner <=