Open Stack

Eduard, Gorka has done a great job of explaining some of the issues with Active-Active Cinder-volume services in his blog:
http://gorka.eguileor.com/

TL;DR: The hacks to use the same hostname or use Pacemaker + VIP are dangerous because of races, and are not recommended for Enterprise deployments.

From: Eduard Matei [mailto:eduard.matei at cloudfounders.com]
Sent: Tuesday, September 15, 2015 8:54 AM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Cinder]Behavior when one cinder-volume service is down

Hi,

Let me see if i got this:
- running 3 (multiple) c-vols won't automatically give you failover
- each c-vol is "master" of a certain number of volumes
-- if the c-vol is "down" then those volumes cannot be managed by another c-vol

What i'm trying to achieve is making sure ANY volume is managed (manageable) by WHICHEVER c-vol is running (and gets the call first) - sort of A/A - so this means i need to look into Pacemaker and virtual-ips, or i should try first the "same name".

Thanks,

Eduard

PS. @Michal: Where are volumes physically in case of your driver? <- similar to ceph, on a distributed object storage service (whose disks can be anywhere even on the same compute host)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150915/d57b43d1/attachment.html>