xfs
[Top] [All Lists]

Re: task blocked for more than 120 seconds

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: task blocked for more than 120 seconds
From: "Josef 'Jeff' Sipek" <jeffpc@xxxxxxxxxxxxxx>
Date: Mon, 23 Apr 2012 13:16:58 -0400
Cc: xfs@xxxxxxxxxxx
In-reply-to: <20120421002932.GG9541@dastard>
References: <20120418151139.GC4652@xxxxxxxxxxxxxxxxxxxxxx> <20120418234821.GR6734@dastard> <20120419154601.GB8230@xxxxxxxxxxxxxxxxxxxxxx> <20120419225603.GA9541@dastard> <20120420135820.GB9189@xxxxxxxxxxxxxxxxxxxxxx> <20120421002932.GG9541@dastard>
User-agent: Mutt/1.5.21 (2010-09-15)
On Sat, Apr 21, 2012 at 10:29:32AM +1000, Dave Chinner wrote:
> On Fri, Apr 20, 2012 at 09:58:20AM -0400, Josef 'Jeff' Sipek wrote:
...
> > Right.  Here's the output from a few minutes ago (I removed the empty lines
> > and redundant column headings):
> 
> Thanks, that makes it easy to read ;)

Yeah, I don't know who's brilliant idea it was to output the headings every
time.

> Now comes the "more data needed" bit. I still need the sysrq-w
> output,

D'oh!  Totally forgot to put that up somewhere.

http://31bits.net/download/xfs_log.gz

Sadly, it looks like the kernel's ring buffer wrapped while dumping the ~180
stacks.

> but also given the length of the incident, some other data
> is definitely needed:
> 
>       - a 30s event trace - it'll compress pretty well
>         (trace-cmd record -e xfs* sleep 30; trace-cmd report > output.txt)

Hrm... Time to figure out how to get ftrace going on Gentoo.  I don't see an
obvious package to install to get trace-cmd.

>       - vmstat 1 output for the same period
>       - output of /proc/meminfo at the same time
>       - the same iostat output for comparison.

Jeff.

-- 
I'm somewhere between geek and normal.
                - Linus Torvalds

<Prev in Thread] Current Thread [Next in Thread>