Post #4437568
2026-04-23 22:23 UTC
@mmeier@social.mei-home.net
Yes the latter applies (I am a former multi-year maintainer, as a disclaimer).
I read the previous posts only briefly and read something about services alongside the main workflow pods.
If disk performance with DBs as services is an issue, you might want to check out tmpfs even, depending on the exact use case.
RWX shouldn't slow down image building as most image builds don't rely heavily on disk IO but rather on CPU. Unless you copy many small files from multiple base images etc.
Most images don't install/copy that much but build a single binary (cpu heavy) and are done.
But yeah, I don't know your image or exact use case so if you're sure it slows down heavily and there isn't an alternative way to build it, then it might just be so.
Background: the way you described the issue in the former posts made it sound to me as if there was a potential misunderstanding on the conceptual level - but I might be wrong ;)
Replies (1)
-
@mmeier@social.mei-home.net 2026-04-23 23:06
@pat_dev@social.linux.pizza No, the issue isn't in the disk perf with the DB service. I haven't gotten to the point where I could even use said service in my workflow. I did extensive testing to arrive at the fact that image builds are slower on CephFS volumes than RBD volumes. If you're interested in the full enchilada, have a look at this post: https://blog.mei-home.net/posts/improving-container-image-build-perf-with-buildah/ I'm still not 100% sure what went wrong there, but using an RBD for the image build was way faster than using CephFS.