[libvirt-users] Pool, iSCSI and guest start

[Dave Allan]

On 03/19/2010 04:44 AM, Nicolas Greneche wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Dave Allan wrote:
>> On 03/17/2010 06:38 AM, Nicolas Greneche wrote:
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> Hi,
>>>
>>> Former user of Xen and newbie in kvm/qemu/libvirt stuff, I give it a try
>>> on my network ;-)
>>>
>>> I need to run a VM with iSCSI target attached.
>>>
>>> I did it this way :
>>>
>>> 1) Creation of iscsi pool (equa.xml) :
>>>
>>> <pool type="iscsi">
>>> <name>equalog</name>
>>> <source>
>>> <host name="10.10.0.1"/>
>>> <device
>>> path="iqn.2001-05.com.equallogic:0-8a0906-4992c7f05-39c000000114b8fc-vglog"/>
>>>
>>> </source>
>>> <target>
>>> <path>/dev/disk/by-path</path>
>>> </target>
>>> </pool>
>>>
>>> This pool start smoothly (when open-iscsi started), no problems. An
>>> entry is created in /dev/disk/by-path/ related to iscsi target.
>>>
>>> 2) I flagged it autostart :
>>>
>>> root at sandi:~# virsh pool-autostart equalog
>>> Pool equalog marked as autostarted
>>>
>>> root at sandi:~# virsh pool-list
>>> Name                 State      Autostart
>>> - -----------------------------------------
>>> equalog              active     yes
>>>
>>> 3) In my guest VM, I have following section :
>>>
>>>       <disk type='block' device='disk'>
>>>         <driver name='qemu'/>
>>>         <source
>>> dev='/dev/disk/by-path/ip-10.10.0.1:3260-iscsi-iqn.2001-05.com.equallogic:0-8a0906-4992c7f05-39c000000114b8fc-vglog-lun-0'/>
>>>
>>>         <target dev='vdc' bus='virtio'/>
>>>         <alias name='virtio2'/>
>>>         <address type='pci' domain='0x0000' bus='0x00' slot='0x06'
>>> function='0x0'/>
>>>       </disk>
>>>
>>> When I start VM, iscsi target is availaible.
>>>
>>> The snag is that when I reboot the host, the pool is not automatically
>>> started (making it impossible to autostart VM relying on this iscsi
>>> volume).
>>>
>>> I verified that open-iscsi is started first. Startup script is localised
>>> in /etc/rcS.d which is prior to /etc/rc2.d (my default runlevel).
>>> Libvirtd is started in rc2.d and not mentionned in rcS.d.
>>>
>>> My questions are :
>>> - - Is this the correct way to attach iscsi volume to a guest ?
>>> - - Did I missed something to have iscsi pool autostart working at boot
>>> time ?
>>
>> You're doing everything right, so it's odd that the pool isn't
>> autostarting.  Does the pool autostart properly if you restart libvirtd
>> when the system is fully booted?
>>
>> Dave
>>
>>
>
> Hi Dave,
>
> It's a very odd problem. Making network debugging with tcpdump makes me
> see that my network stack doesn't receive "arp reply" related to my target.
>
> If I add an ARP entry by hand in cache or a sleep just before libvirtd
> start function in startup script it works like a charm.
>
> Very odd, I asked debian package maintainers for help :
>
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=574358

Are the appropriate network interfaces started/configured?