Re: infinite loop when doing disk zone testing

From: Jens Axboe <jens.axboe_at_oracle.com>
Date: Thu, 31 Jan 2008 14:25:33 +0100

On Thu, Jan 31 2008, Ryan Thomas wrote:
> Actually I've also replicated this behavior with with 1.17.2 with the
> iodepth patch as well.

please try the current version:

http://brick.kernel.dk/snaps/fio-git-latest.tar.gz

(and pretty please stop top posting when you reply to emails!

>
> -----Original Message-----
> From: Ryan Thomas
> Sent: Thursday, January 31, 2008 8:23 AM
> To: 'Jens Axboe'
> Cc: fio-devel_at_kernel.dk
> Subject: RE: infinite loop when doing disk zone testing
>
> Sorry, 1.17 with the iodepth patch for zone alignment you just posted.
>
> -----Original Message-----
> From: Jens Axboe [mailto:jens.axboe_at_oracle.com]
> Sent: Thursday, January 31, 2008 8:14 AM
> To: Ryan Thomas
> Cc: fio-devel_at_kernel.dk
> Subject: Re: infinite loop when doing disk zone testing
>
> On Thu, Jan 31 2008, Ryan Thomas wrote:
> > I have a disk that is 976773117 sectors (512 bytes) in size. I tried
> > running a disk zone test with the following parameters
> >
> >
> >
> > [global]
> >
> > bs=64k
> >
> > direct=1
> >
> > rw=read
> >
> > ioengine=libaio
> >
> > iodepth=2
> >
> > zonesize=256m
> >
> > zoneskip=2g
> >
> > write_bw_log
> >
> >
> >
> > [/dev/e214.0]
> >
> > write_iolog=foo2
> >
> >
> >
> > When I try to run this job, everything goes well until the job reaches
> > 90% done and at that point the io goes quiet. At that point the iolog
> > shows that fio goes into an infinite loop of opening and closing the
> > device.
> >
> >
> >
> > /dev/e214.0 read 500107640832 65536
> >
> > /dev/e214.0 read 500107706368 65536
> >
> > /dev/e214.0 close
> >
> > /dev/e214.0 open
> >
> > /dev/e214.0 close
> >
> > /dev/e214.0 open
> >
> > /dev/e214.0 close
> >
> > /dev/e214.0 open
> >
> > /dev/e214.0 close
> >
> > /dev/e214.0 open
> >
> > /dev/e214.0 close
> >
> > ....
> >
> >
> >
> >
> >
> > If I run a slightly different job where I try to align the zones at 2g
> > boundaries by specifying zoneskip=1792m (g-256m ) the job runs just
> > fine.
> >
> >
> >
> > The problem is that the disk only has 207 full zonesize plus zoneskip
> > cycles. The partial zonesize range at the end falls of the end of the
> > disk resulting in failed io and apparently fio is retrying the device
> to
> > try and finish that last zone.
>
> Which fio version?
>
> --
> Jens Axboe

-- 
Jens Axboe
Received on Thu Jan 31 2008 - 14:25:33 CET

This archive was generated by hypermail 2.2.0 : Thu Jan 31 2008 - 14:30:02 CET