[vUDS Blueprint 13.10 ]OpenStack HA Next Steps

[Andres Rodriguez]

Dear all,

This past cycle Ubuntu Server members worked very hard to provide a
Reference Architecture for OpenStack HA, largely based on technologies such
as Pacemaker and Corosync. [1]. The main work, besides defining the
architecture, was concentrated on implementing a hacluster charm, as well
as adding HA support for all of the related OpenStack Charms. This is what
we consider HA v1.

Now that the new cycle has begun, we would like to start working towards a
HAv2 reference architecture. However, we need your input in the following
topics.

 * MySQL HA. The current HA implementation for MySQL is Active/Passive,
using Ceph as a storage back-end. However, we would like to be able to use
an Active/Active scenario  For this reason, we are considering the use of
Percona/MariaDB and the use of Galera cluster, which provides this
functionality.

 * RabbitMQ: The sample applies to RabbitMQ. We are currently using an
Active/Passive implementation using Ceph as a back-end as well. We would
like to discuss the possibility of moving away from RabbitMQ in favor of
ZeroMQ, which provides HA out of the box.

* HA Awareness  We would also like to start discussing how much knowledge
of the service availability should the HA implementation know about.
Currently, we only check that the services have been started and are
running. We would like to provide more in depth monitoring to determine
that not only the services are running, but they are also working properly.

* Subordinate Charm: Currently, the hacluster charm is a subordinate charm
that is related to all of the services that require HA features. We are
looking to improve this and maybe make it available in the common openstack
library we use in the charms.

* Newer Pacemaker/Corosync version. We are evaluating the possibility of
upgrading the version of this software to the latest available (Pacemaker
1.1.9, Corosync 2.0.x). This new upstream releases provide improved
performance for 8/16 node clusters, which helps with the scalability of our
HA implementation for OpenStack. Even though this is unrelated, the only
inconvenient will be that we will be dropping support for OCFS2 clusters.

If you would like to know more about our implementation, please refer to
the wikipage in [1]. We would like to hear from you in the topics above, as
well as feedback with the reference architecture.

Cheers.


[1]: https://wiki.ubuntu.com/ServerTeam/OpenStackHA


-- 
Andres Rodriguez (RoAkSoAx)
Ubuntu Server Developer
Systems Engineer
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/ubuntu-server/attachments/20130502/b9af87c7/attachment.html>