Obstacles for OpenStack: disk_format is not the same as disk_format

I bet you’re wondering about the title and if there’s a typo or some other mistake. I promise, there’s nothing wrong with the title. It’s a result of some research I did in two different environments (Ocata and Rocky), I already wrote an article about some of the findings.

But there was one more thing I couldn’t wrap my head around: although I had changed the disk_format to “raw” in all Glance images I still saw some “qcow2” images from time to time, but I had no idea how they were created. I tried different Nova settings regarding snapshots, but the default already is to upload snapshots in the source’s image format, so no luck with that.
That’s why it took some time, I got lucky because a co-worker reported an issue that was completely unrelated to this. But during my tests to recreate his problem I stumbled upon the disk_format (again).

I had a trace, finally!

And the solution is pretty simple, although I don’t understand the intention (yet). It seems as if snapshots of volume backed instances are uploaded with disk_format = qcow2. This doesn’t make much sense to me, but I could deal with it if there wasn’t the issue with “qcow2” in a Ceph cluster.

So I assumed that launching an instance from this new base image would cause slow boot processes as reported, only it didn’t. Although the base image also shows disk_format = qcow2 as in the previous case it does not have the same effect. Instead the new instance is also launched from a volume that is a rbd clone of the original instance (also volume backed), so in the end it’s the expected workflow, only the image’s disk_format doesn’t make sense here. I hope my description does, though.

If you need some examples in this article, just comment and I’ll add some terminal output. The intention of this post was to simply make you aware that the exact same image property can either cause some headache because your instances boot very slowly, or it doesn’t affect the processes in any way at all. I reported a bug to address this inconsistency.

This entry was posted in Ceph, OpenStack, SUSE Cloud. Bookmark the permalink.

Leave a Reply