[Top] [All Lists]

Re: Locking problems

To: Steve Lord <lord@xxxxxxx>
Subject: Re: Locking problems
From: Chris Tooley <ctooley@xxxxxxxx>
Date: 30 Aug 2002 09:18:14 -0500
Cc: linux-xfs@xxxxxxxxxxx, David Duncan <dduncan@xxxxxxxxxxxxxx>
In-reply-to: <1030715530.9148.160.camel@xxxxxxxxxxxxxxxxxxxx>
References: <Pine.LNX.4.44.0208281747390.7497-100000@xxxxxxxxxxxxxxxxxxxxxx> <1030711109.9300.7.camel@xxxxxxxxxxxxxxx> <1030715530.9148.160.camel@xxxxxxxxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
Hmmm, it may not be file locking.  I'm not sure.  It could be the
ordered access you were talking about but I'm not sure that would get
the same kind of corruption that we're seeing.  I'll pull the data over
to a box with XFS and run with strace?

I'm not exactly sure how to test these things.

Chris Tooley

On Fri, 2002-08-30 at 08:52, Steve Lord wrote:
> On Fri, 2002-08-30 at 07:38, Chris Tooley wrote:
> > 
> > Sorry, was out yesterday.  The corruption we saw was in the data for the
> > application.  It's a point of sale application (which I can provide
> > access to or simply provide to if someone can help me with it) that uses
> > an ISAM database.  It appears that 2 or more people are getting the
> > write access to the same record at the same time.  This doesn't happen
> > on ext2, but does on XFS.  It would appear to me, and I'll admit I've
> > had no sleep for about a week, that it's a file locking issue.  In fact
> > we're seeing problems where two people will get the same ticket number
> > for their sales.  This thing gets ticket numbers by autogenerating a
> > table row for it and passing the key to the client application.  For two
> > applications to get the same ticket number they have to look and see
> > what the next available number is at the same time.  One of them is
> > supposed to lock that file during it's write, but that isn't happening.
> Can you confirm for sure that it is using file locking to do this,
> saying it works for one filesystem and not another is a little
> bizarre since the file locking implementation is above the filesystems
> not inside them. A slightly harder question would be is this mandatory
> file locking or not?
> Having the issue narrowed down like this is a great help in knowing
> where to look.
> Steve
> -- 
> Steve Lord                                      voice: +1-651-683-3511
> Principal Engineer, Filesystem Software         email: lord@xxxxxxx

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