xfs
[Top] [All Lists]

Re: xfs recovery oops in 2.6.4-mm1

To: Nathan Scott <nathans@xxxxxxx>
Subject: Re: xfs recovery oops in 2.6.4-mm1
From: William Lee Irwin III <wli@xxxxxxxxxxxxxx>
Date: Sun, 14 Mar 2004 16:02:44 -0800
Cc: Steve Lord <lord@xxxxxxx>, tes@xxxxxxx, linux-xfs@xxxxxxxxxxx
In-reply-to: <20040314235119.GB643@frodo>
Organization: The Domain of Holomorphy
References: <20040312100025.GP655@xxxxxxxxxxxxxx> <4051D517.4070005@xxxxxxx> <20040312233153.GT655@xxxxxxxxxxxxxx> <4052834B.5010208@xxxxxxx> <20040314233054.GA643@frodo> <20040314233604.GV655@xxxxxxxxxxxxxx> <20040314235119.GB643@frodo>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.5.1+cvs20040105i
On Sun, Mar 14, 2004 at 03:36:04PM -0800, William Lee Irwin III wrote:
>> I created it first with 2.6.3-mm4.

On Mon, Mar 15, 2004 at 10:51:19AM +1100, Nathan Scott wrote:
> ok, thanks.  did you see these recovery problems with that
> kernel too, or is it specific to this later one?

It happened there, too. I pulled the plug during a 2.6.4-mm1 run,
2.6.4-mm1 went down on the way back up, I pulled the plug again
and booted 2.6.3-mm4, and that went down during recovery too.


On Sun, Mar 14, 2004 at 03:36:04PM -0800, William Lee Irwin III wrote:
>> Well, it sort of crapped itself in the midst of doing large
>> recovery ops, so I think marking it dirty and replaying nothing
>> won't fly,

On Mon, Mar 15, 2004 at 10:51:19AM +1100, Nathan Scott wrote:
> thats not quite what this does - the forced shutdown will cause
> a real filesystem recovery to run on the next mount.  try with
> -f too to flush the log (but not metadata) during shutdown.

Okay, I can try that out too.


On Sun, Mar 14, 2004 at 03:36:04PM -0800, William Lee Irwin III wrote:
>> but pulling the plug in the midst of fsstress sounds
>> like a good bet.

On Mon, Mar 15, 2004 at 10:51:19AM +1100, Nathan Scott wrote:
> thats a rather painful way to do testing :).

Well, I have tried pulling the plug while the fs is idle since, but
that doesn't trigger it during recovery. When I triggered this, an
app that put about 8000 16MB-sized (dio) aio's in flight was in
progress. That looks like too slow a way to do it since I waited a
while before pulling the plug (impatience was why I tried it), but
afterward, pulling the plug too soon didn't trigger it either. I only
got it to happen again if I waited about an hour into that before
pulling the plug on the stuff.


-- wli


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