Re: I/O scheduler deadlocks on Xen virtual block devices

[Jeff Moyer <jmoyer@redhat.com>]

Doug Neal <dneallkml@gmail.com> writes:

>>>>
>>>> Did you try these different I/O schedulers in the domU or on the dom0?
>>>> Does switching I/O schedulers in either place make the problem go away
>>>> when it happens?
>>>>
>>>
>>> In the domU, and the bug was present in all cases. The dom0 was using
>>> cfq. I'll run the tests again using each scheduler in the dom0 with
>>> domU set to noop and report back.
>>
>> While I think this is an interesting test, you need only test one I/O
>> scheduler in the dom0.  Also, I think you misunderstood the second
>> question.  I'd like to know if switching I/O schedulers while the system
>> is in this bad state helps get I/O going again.
>>
> Jeff,
>
> Issue confirmed in the following cases:
>
> dom0:domU
> noop:noop
> noop:deadline
> noop:anticipatory
> cfq:noop
> cfq:deadline
> cfq:anticipatory
> cfq:cfq
>
> I try switching schedulers in the domU during the lockup, the echo x >
> /sys/block/xvda/queue/scheduler command also hangs. In the dom0, no
> effect.

OK.  So, this means that the I/O is not hung up in the I/O scheduler.  I
don't have a whole lot of experience with the Xen I/O stack, but I think
that's where I'd start digging (especially if there's nothing in the
logs about command timed out).

I hope this was a somewhat helpful exercise.  Sorry I can't be of more
help.

Cheers!
Jeff