kubedirector
kubedirector copied to clipboard
revisit persistent storage model
In K8s land, persistent shared storage is needed not only for migration, but also if we want to support a pause/restart kind of behavior that doesn't lose disk state.
I think we have some leftover "old thinking" about persistent storage that needs to be revisited. Questions like:
- [x] Should we really have default directories or always explicitly list them all? (answer: defaults are good)
- [x] App CR should probably should be allowed to specify the directories per-role.
- [x] Can/should we support directories that are not top-level dirs?
- [ ] How "optional" is persistent storage really; what is the correct interaction between site default, app requirements, and deploy-time choices?
May require issue #5 (KD site settings).
all default directories /etc, /opt, /var and /usr are required for bdvcli and appconfig to work after migration. we can move all the guestconfig related files to a single folder.
How "optional" is persistent storage really; what is the correct interaction between site default, app requirements, and deploy-time choices?
- Site Default must specify one or more storaClass names that is setup (we don't have this today)
- App requirements should indicate which directories are required for a successful migration
- deployment time should purely indicate which storageClass and size to use.
BTW @riteshja has signed up to look at the KD administrative configuration stuff (issue #5).
Moving to 0.3 for the remaining design issues.