Description of problem: The "Getting Started" guide in "16.2. Attaching a Volume using the Command Line Interface" describes: "" $ nova volume-attach INSTANCE_ID VOLUME_ID auto The auto parameter indicates that Nova must attempt to automatically assign assign a device identifier to the volume within the guest. Alternatively you may specify a device identifier such as /dev/vdb. "" The latter part of this needs more explanation because it seems that providing /dev/vdb may not actually assign that device node to the disk -- it seems to behave as if you have specified "auto". This is unfortunate since the device node is a required parameter for volume-attach. https://bugs.launchpad.net/nova/+bug/1004328 has some info on this as does http://www.mail-archive.com/openstack@lists.launchpad.net/msg15593.html I think we need to document an explanation of how/if this works for RHOS.
* /dev/vdb does not exist: $ nova volume-attach INSTANCE_ID VOLUME_ID auto -> volume is set to /dev/vdb $ nova volume-attach INSTANCE_ID VOLUME_ID/dev/vdx -> volume is set to /dev/vdb * /dev/vdb exists: $ nova volume-attach INSTANCE_ID VOLUME_ID auto -> volume is set to /dev/vdc $ nova volume-attach INSTANCE_ID VOLUME_ID /dev/vdx -> volume is set to /dev/vdc The main issue is that horizon dashboard and cli tools report the command line user choice while the volume is set to default.
I updated the statement to read "The <parameter>auto</parameter> parameter indicates that Nova must attempt to automatically assign assign a device identifier to the volume within the guest. Manual allocation of specific device identifiers within the guest is not supported on KVM hypervisors at this time." The fact that the client tools (incl. the dashboard) will report whatever the user enters for this parameter even though it is never applied (except where it is coincidentally the same as what auto would select) is all the more reason to tell them not to use anything but auto in my opinion.
Moving back to ASSIGNED, still need to consider a second patch to highlight the use of disk-by-id to identify the volumes. Seems more reliable than the advice we have in the guide currently, even when using auto.
commit dae53f6d79e4902561a81efae49908d867620525 Author: Stephen Gordon <sgordon> Date: Wed Mar 13 14:34:16 2013 -0400 BZ#918809 - Use /dev/disk/by-id/ to access volumes. Attached volumes are now exposed to guests under both the traditional device tree (/dev/vd*) and by ID (/dev/disk/by-id/). Using the latter volumes attached to Nova compute instances now expose part of their Cinder ID, allowing users to uniquely identify them from inside the guest. The information on accessing volumes from inside instances has been updated to reflect that this is a more desirable way of referring to the instances. The example for formatting a volume has been updated to access the volume in this way and the IDs used have been replicated to the rest of the content to ensure consistency. Change-Id: If69553c911a92cea0384e634f5df6a4295fe734d