Helm: Configurable OSD directory or devices#562
Conversation
|
Do you mind rebasing this on top of master please? |
| osd_directory: | ||
| enabled: true | ||
| osd_devices: | ||
| - device: sdb |
There was a problem hiding this comment.
This is an issue for people using persistent path such as /dev/disk/by-.
Can we ask for the full path instead?
There was a problem hiding this comment.
Yes, that is the downside with this approach and one of the reasons why I'd opened the PR for discussion.
We use the device for building the name and selectors in the daemonset (e.g. name: ceph-osd-{{ $value.device }}-{{ $value.type }}) so won't work for the longer by-* paths. I haven't really found with a decent alternative that will work for all cases. Any thoughts?
There was a problem hiding this comment.
Alright, so I've been thinking and put my thoughts into this PR, would you mind having a look at the README and tell me what you think? I left some questions too. Thanks!
#610
The PR is obviously incomplete at the moment but the general idea is to keep one OSD per container and not to have to declare an OSD device for it. In an ideal world someone would ask k8s, deploy me a Ceph cluster from a specific set of labeled machines (storage node), take all the disks and use them to build my Ceph cluster.
|
I'm closing this since this should probably go in https://github.com/ceph/ceph-helm/. |
2 participants
I thought I'd raise this PR to get some discussion going on the best way to handle multiple OSDs in a cluster. There has been some discussion in various places about real disk support.
We've been using an approach with daemonsets per device across a cluster. It gives more control to the OSD lifecycle with drive types, crush location and rolling restarts. Its served us reasonable well for clusters with identical servers/disks but may be difficult to manage in a varied cluster.
It may be that the best solution is a ThirdPartyResources/Operator but for now this does allow the use of real devices with Helm. Please jump in with thoughts, alternatives and ideas