Restart of nova-volume with volumes in wrong state: global name 'volume_ref' is not defined

[Christian Berendt]

I tried using the SheepdogDriver by setting the flag "--volume_driver=nova.volume.driver.SheepdogDriver" in /etc/nova/nova.conf. Now I try to start nova-volume, but I end in this error:

2011-02-05 00:32:35,834 AUDIT nova.root [-] Starting volume node (version 2011.1-workspace:tarmac-20110202111800-10sib5zxlu8xluca)
2011-02-05 00:32:36,088 CRITICAL nova.root [-] global name 'volume_ref' is not defined
(nova.root): TRACE: Traceback (most recent call last):
(nova.root): TRACE: File "/usr/bin/nova-volume", line 44, in <module>
(nova.root): TRACE: service.serve()
(nova.root): TRACE: File "/usr/lib64/python2.6/site-packages/nova/service.py", line 231, in serve
(nova.root): TRACE: x.start()
(nova.root): TRACE: File "/usr/lib64/python2.6/site-packages/nova/service.py", line 81, in start
(nova.root): TRACE: self.manager.init_host()
(nova.root): TRACE: File "/usr/lib64/python2.6/site-packages/nova/volume/manager.py", line 90, in init_host
(nova.root): TRACE: LOG.info(_("volume %s: skipping export"), volume_ref['name'])
(nova.root): TRACE: NameError: global name 'volume_ref' is not defined
(nova.root): TRACE:

[Christian Berendt (berendt)]

problem is in init_host() (nova/volume/manager.py) in line 89-90. volume_ref is not set in this method, but the log message wants to use it. i simply uncommented the both lines and now i can start nova-volume without an error. Now I'll test if I can use Sheepdog images...

(I hope I added the patch in the correct way...)

[Christian Berendt (berendt)]

I think it looks good now...

# euca-create-volume -s 5 -z nova
VOLUME vol-00000004 5 creating (berendt, None, None, None) 2011-02-05 17:49:23.394126

# euca-describe-volumes
[...]
VOLUME vol-00000004 5 nova available (berendt, mandarin, None, None) 2011-02-05 17:49:23

# collie vdi list
  name id size used shared creation time vdi id
------------------------------------------------------------------
  volume-00000004 1 5.0 GB 0.0 MB 0.0 MB 2011-02-05 18:49 a1ff92

# collie vdi object volume-00000004
Looking for the inode object 0xa1ff92 with 4 nodes

172.19.134.24:7000: doesn't have
172.19.134.26:7000: has the object (should be 2 copies)
172.19.134.25:7000: has the object (should be 2 copies)
172.19.134.27:7000: doesn't have

[Vish Ishaya (vishvananda)]

This patch affects all volume modes if volume is started with a volume that isn't in available or in-use

[Vish Ishaya (vishvananda)]

The patch above isn't quite correct, volume_ref should just be changed to volume.

[Thierry Carrez (ttx)]

I'm looking into validating that this is indeed fixed in 2011.1.1 release candidates... but I'm not sure I understand how to reproduce the bug (or absence thereof). Can anyone involved come up with a simple test case that I could use in validation ?

[Thierry Carrez (ttx)]

validated:

with bzr645:
CRITICAL nova.root [-] global name 'volume_ref' is not defined

with bzr653:
Re-exporting 1 volumes from MainProcess (pid=7167) init_host
INFO nova.volume.manager [-] volume volume-00000002: skipping export