High Availability from the DevOps side - OpenStack Summit Portland

High Availability in OpenStack

from DevOops! side

Emilien Macchi, Sébastien HanCloud Engineers @ eNovance

whoarewe

Seb

- DevOps- Ceph Specialist- Blogger

Emile

- DevOps- Doc Contributor- Quantum Fanatic

ContributionsPacemaker Resource AgentsOfficial DocumentationPuppet Modules

ExperiencesMedium Public CloudLarge Public Cloud

Look further

Agenda

Contributions

- Essex / Folsom / Grizzly- Covers SPOF services- Future: no future

Pacemaker Resource Agents

Download them all!https://github.com/madkiss/openstack-resource-agents

- Active / Passive mode- MySQL- RabbitMQ- OpenStack APIs- Need contributors- Future: Active / Active

Official Documentation

RTFM !http://docs.openstack.org/trunk/openstack-ha/

- Update modules for Grizzly - Nova | Cinder | Quantum are HA aware

- Make modules for HA deployments (Work in Progress)

Puppet Labs

Experiences

Use case #1: Medium-Size Public Cloud

coming soon

✓ Easy and fast deployment (with Puppet)✓ Enough for our needs

✗ Not scalable enough✗ Split brains risks✗ DRBD?

Can be improved:Migrate to Grizzly, change for active / active...

Feedback

Use case #2: Large-Size Public Cloud

✓ Highly scalable✓ Maintenance✓ The Network scheduler allows us to scale L3

✗ Networking: L3 is not HA✗ OVS plugin does not scale (L2 isolation)

In progress : testing cells

Feedback

To infinity... and beyond

- Nova-conductor (horizontal scaling)- Ceilometer improvements- Quantum Load Balancing & Metadata agents

Newcomers

Des questions ?

E-mail : emilien@enovance.comIRC : EmilienMTwitter : @EmilienMacchiBlog : http://my1.fr/blog

E-mail : sebastien@enovance.comIRC : lesebTwitter : @sebastien_hanBlog: http://sebastien-han.fr/blog