GlusterFS is Ready for OpenStack

Amidst the madness of the OpenStack Summit a couple of weeks ago, you could be forgiven for not seeing a Red Hat announcement about GlusterFS being “OpenStack-ready”. You may wonder, what exactly do we mean by “OpenStack-ready”?

The first thing to understand is that storage in OpenStack is multifaceted. It could mean applications storing objects over HTTP in Swift, OpenStack’s object storage implementation. It could mean the management of VM images with glance, which stores VMs and tracks their metadata in a catalog. Or it could mean the Cinder interface for Nova, which gives OpenStack operators an interface to block storage devices, real or virtual.

When we say that GlusterFS is “OpenStack-ready” we mean that GlusterFS can be used for four use cases, today:

– The Gluster-Swift implementation: enables GlusterFS to act as the storage back-end to the Swift API and proxy server.
– Glance image catalog: Glance can store images on GlusterFS when setting the storage backend to “file.”
– Gluster-Cinder integration: Beginning with OpenStack Grizzly, the Cinder interface now supports GlusterFS directly, meaning that OpenStack operators can more easily deploy and access VMs on mounted GlusterFS volumes.
– Instance storage: running VM images on a shared volume to facilitate live migration

The maturity and scalability of each of these differs. The Swift implementation is, relatively speaking, the most mature and has been deployed successfully at some sites in production. Using GlusterFS to back the image storage for Glance is also mature as it requires no code changes to Glance. The Cinder interface is quite new. Anyone who expresses an interest in using GlusterFS with Cinder should be forewarned that this code is still in its infancy – begun in February and completed in March. Instance storage is something that, while possible and should work, may not scale well or perform up to the expectations of some operators until future releases.

For now, Swift and Glance support are mature enough that they are readily integrated into many OpenStack environments, including those based on Folsom. Block storage via the Cinder interface is something that is possible only for Grizzly-based OpenStack deployments. Many of these features will be improved, performance and feature-wise, when GlusterFS 3.4 is released with the QEMU + libgfapi integration. Right now, the release date looks to be in early June, with beta 1 due to drop today (May 7).

So when we say GlusterFS is ready for OpenStack, we mean it comprehensively, and it will only get better with the added QEMU integration.