Re: [Qemu-devel] rbd: Possible regression in 2.9 RCs

[Alexandru Avadanii]

Hi, Eric,
Thank you for looking into this!

> -----Original Message-----
> From: Eric Blake [mailto:address@hidden
> Sent: Friday, March 31, 2017 3:40 AM
> To: Alexandru Avadanii; address@hidden
> Cc: svc-armband; Jeff Cody; Markus Armbruster
> Subject: Re: [Qemu-devel] rbd: Possible regression in 2.9 RCs
> 
> On 03/30/2017 06:42 PM, Alexandru Avadanii wrote:
> > Hi,
> > While testing out 2.9.0-rc2 on AArch64, we noticed a possible regression in
> QEMU, related to parsing -drive 'file=rbd/...':
> >> "conf option 6789 has no value".
> >
> > Instance logs [1].
> 
> Pastebins don't last forever; it helps to paste the actual error message in 
> the
> email for archival purposes, and to make it easier for readers to see your
> problem without having to chase URLs:
> 
> 2017-03-30T20:02:27.499695Z qemu-system-aarch64: -drive
> file=rbd:volumes/volume-ea141b5c-cdb3-4765-910d-
> e7008b209a70:id=compute:key=AQAVkvxXAAAAABAA9ZxWFYdRmV+DSwKr
> 7BKKXg==:auth_supported=cephx\;none:mon_host=192.168.1.2\:6789,form
> at=raw,if=none,id=drive-virtio-disk0,serial=ea141b5c-cdb3-4765-910d-
> e7008b209a70,cache=writeback:
> conf option 6789 has no value
> 
> Looking at that, the only instance of 6789 that I see is the
> 'mon_host=192.168.1.2\:6789,' portion.  I bet what is happening is that we
> are mis-parsing the string, and trying to treat it as a key-values pair.  In 
> other
> words, it's probably an unintended regression introduced in the range of
> 7830f909..0a55679b by Jeff [3] or in Markus' cleanups between
> f51c363c..2836284d [4].
> 
> On the bright side, we still have time to fix it before 2.9 goes final, now 
> that
> you called it to our attention.
> 
> > Occasionally, we get "conf option too long", with the same effect.
> >
> > We bisected this manually between 2.8.0 (working ok with the above cmd)
> and 2.9.0-rc2, and the problematic change seems to be the merge point [2].
> 
> I suspect you didn't run the bisect quite correctly, as that merge point has
> nothing to do with block/rbd.c.

I suspect that too, sorry. I'll redo this tomorrow and get back.

> 
> >
> > [1] http://paste.openstack.org/show/604938/
> > [2] https://github.com/qemu/qemu/commit/9a81b79
> >
> 
> [3] https://lists.gnu.org/archive/html/qemu-devel/2017-02/msg07506.html
> [4] https://lists.gnu.org/archive/html/qemu-devel/2017-03/msg05565.html
> 
> --
> Eric Blake   eblake redhat com    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org