I finally completed a long due project at work. After virtualizing a large number of systems, and moving them under a single management interface, it was time for high availability. But, it had to be done one a budget. And, by budget, I mean, almost nothing.
For virtualization we already utilize Proxmox which is open source. It may not be as feature rich as some commercial products, but it is reliable. Naturally, since we’re already using open source virtualization, we might as well use open source storage. Since I was repurposing existing equipment, the decision came down to DRBD or Gluster.
But, which to go with? I tested out both on a mix of systems. Basically, I used whatever I could get my hands on. DRBD showed a lot of promise, and the integration with Proxmox was nice. Unfortunately, I ran into split brain problems several times. That left me with Gluster, which so far hasn’t suffered from that same fate. Though, to lessen the odds of that, we do run in a 3-way replica configuration.
So, how’s it all implemented? It’s a hyper-converged setup. The VM hosts double as storage hosts. Our typical server is a 2U box, as most are dual and quad socket systems. In addition, they all have dual power supplies. So, they have plenty of room for additional disks. We’re running multiple volumes spread across our nodes to limit the amount of healing during maintenance.
Of course, it’s only been in production a short period of time. So, time will tell. And, if it all goes side ways, I’m sure to rant about it. Until then, educate yourself on distributed storage.