Bug 1062693

Summary:	[RFE]foreman should config nova for libgfapi access to cinder volumes
Product:	Red Hat OpenStack	Reporter:	Steve Reichard <sreichar>
Component:	openstack-foreman-installer	Assignee:	John Eckersberg <jeckersb>
Status:	CLOSED WONTFIX	QA Contact:	nlevinki <nlevinki>
Severity:	urgent	Docs Contact:
Priority:	high
Version:	4.0	CC:	aberezin, cpelland, jeckersb, morazi, rhos-maint, scohen, srevivo, ssaha, yeylon
Target Milestone:	async	Keywords:	FutureFeature, ZStream
Target Release:	4.0
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	Enhancement
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2014-09-08 16:38:00 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:	1020483
Bug Blocks:	1040649, 1045196

Description Steve Reichard 2014-02-07 18:00:39 UTC

Description of problem:


With RHEL 6.5 cinder can be accesses using libgfapi, which is my understanding provides better performance and supports the volume snapshot features.

relates to packstack BZ 1020483

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info:

Comment 2 Martin Magr 2014-03-27 11:24:46 UTC

*** Bug 1020483 has been marked as a duplicate of this bug. ***

Comment 3 John Eckersberg 2014-03-27 19:38:03 UTC

Changed title to reflect that nova, not cinder, should use libgfapi.

Cinder does not support using libgfapi as far as I can tell.  The class cinder.volumes.drivers.GlusterfsDriver explicitly requires mount.glusterfs and uses the fuse layer to interface with gluster.

Foreman already has all the necessary knobs exposed to use cinder in this way.  Set cinder_backend_gluster to true, and then customize cinder_gluster_servers and cinder_gluster_volume for the environment.  I've verified that I can point this at a dummy gluster server and successfully do 'cinder create 1' to create a test volume.

Nova however, via libvirt/qemu, can use libgfapi to access gluster on the compute nodes.  At a glance, that does not appear configurable in foreman.  I'm investigating that piece more now.

Comment 4 Steve Reichard 2014-03-28 16:40:58 UTC

not sure that title is accurate either. 

it is using libgfapi | libvirtd | nova for cinder volumes.

At this point we can not make nova use libgfapi for instance/ephemeral storage.

Comment 5 John Eckersberg 2014-03-28 16:53:00 UTC

Review for puppet-nova adding required support - https://review.openstack.org/83816

Comment 6 John Eckersberg 2014-03-28 18:17:40 UTC

The above patch was rejected.  There's a new nova::config class that is more appropriate for this, so I'll see what's involved to backport it for havana.

Comment 7 John Eckersberg 2014-03-28 18:48:45 UTC

New plan, instead of going through the hassle of backporting nova::config, we can just create nova_config resources directly in quickstack::compute_common when cinder_backend_gluster is set.

Also, we'll need to make sure glusterfs-api is installed on the compute nodes so that libgfapi is available for qemu.  I've submitted a pull request for that bit here - https://github.com/redhat-openstack/puppet-openstack-storage/pull/5

Comment 8 John Eckersberg 2014-04-15 16:39:55 UTC

PR - https://github.com/redhat-openstack/astapor/pull/167

Comment 9 John Eckersberg 2014-04-16 20:30:05 UTC

If there is hope of this ever actually working end to end, these two bugs will need to be fixed:

https://bugzilla.redhat.com/show_bug.cgi?id=1088589
https://bugs.launchpad.net/qemu/+bug/1308542

Particularly the first one.  If that one is fixed it becomes a lot more difficult to fall into the error case of the second bug.