iscsi client hangs performing I/O on a dead target

Angelo nglrossi at gmail.com
Wed Jan 7 17:17:35 UTC 2009


Danny,

this sounds great (except for the panic, but I guess this would at least be
the same behaviour of two scsi disks configured in raid1 with one failing).

I'll keep an eye on the changes and be glad to help testing if you implement
the timeout setting!

thanks,
Angelo

On Wed, Jan 7, 2009 at 2:27 PM, Danny Braniss <danny at cs.huji.ac.il> wrote:

> > ------=_Part_16620_23870833.1231325237102
> > Content-Type: text/plain; charset=ISO-8859-1
> > Content-Transfer-Encoding: 7bit
> > Content-Disposition: inline
> >
> > On Wed, Jan 7, 2009 at 10:33 AM, Danny Braniss <danny at cs.huji.ac.il>
> wrote:
> >
> > > > Hi,
> > > >
> > > > I'm experimenting on iscsi; both the target and the initiator are
> vmware
> > > > virtual machines and are running 7.0-RELEASE.
> > > > So far everything works fine except that the client hangs forever
> (state
> > > is
> > > > PHYSWR) if it is accessing (dd if=/dev/zero of=/dev/da0) the device
> while
> > > it
> > > > "dies" (shutdown of the target machine).
> > > >
> > > what happens when you bring back the target?
> > >
> > > > Is this the expected behaviour?
> > > >
> > > yes.
> > > but maybe it can be 'fixed', I'm concidering adding some kind of
> timeout
> > > flag/option.
> > >
> > > > I would like to be able to configure a timeout so the command that is
> > > trying
> > > > to access the unavailable device fails and I can handle the problem
> but I
> > > > can't find anything in iscontrol man pages.
> > > >
>
> > > > Thanks,
> > > > Angelo
> > > > _______________________________________________
> > > > freebsd-scsi at freebsd.org mailing list
> > > > http://lists.freebsd.org/mailman/listinfo/freebsd-scsi
> > > > To unsubscribe, send any mail to "
> freebsd-scsi-unsubscribe at freebsd.org"
> > > >
> > >
> > > Hi Danny,
> >
> >
> > thanks for your reply.
> >
> > when the target comes back it is reconnected properly by the initiator.
> >
> > The usage of iscsi I had in mind was an initiator connecting to two
> targets
> > and using the two devices as parts of a raid 1 array.
> > I was expecting that a failure on one of the disks (network down, target
> > machine down) would be eventually detected and the raid turned to
> degraded
> > but still usable.
> > What happens now to me is that the whole raid device gets unusable for
> all
> > the time the target is down.
> >
> > As I said before I am just experimenting. I don't know how other
> > implementations of iscsi initiator work and maybe my idea is just insane
> :)
> >
> I don't know either, the RFC does not mention this case - or maybe I missed
> it
> :-)
> But others have raised the issue, and so it got me thinking, but freebsd
> likes
> to
> panic if a disc goes away, so don't hold your breath for a solution soon.
>
> danny
>
> > Thanks,
> > Angelo
>
>


More information about the freebsd-scsi mailing list