Last week was the OpenStack Juno Design Summit in Atlanta Georgia. I spent a lot of time in the Operations track, and there was a recurring theme: “OpenStack needs more enterprise features”. OKay, fair enough. But what features? “Integration into enterprise authentication?” Sure, we do that now with SAML2 in Keystone. “HA of OpenStack services” Also, available now. Two different methods even! You can run active/passive with corosync, or (my preferred way) run mutliple instances behind a highly-available load balancer. But now we get into rest of the HA story and these are where I have a problem:
- “HA and durability of VMs. If a VM crashes, OpenStack should restart it.”
- “A distributed filesystem/NAS solution managed by OpenStack.”
- “Restricting VMs to specific hosts.”
- “Automatic recovery of crashed hypervisors.”
To these types of questions I have a simple answer: No. We’re not VMware. Cloud instances are not pets, or special snow flakes. They are disposable. If you are looking for the features of VMware but don’t want to pay their prices, check out Xen or oVirt. Those are the VMware competitors. OpenStack is different. It’s going to take development changes to you application. It’s going to take a different way of thinking. All of these are good things. Bringing OpenStack up to feature parity of VMware will take too long, and we’re already chasing Amazon. We don’t need to be chasing VMware too when other options exist.