Ceph » CephFS

Well, we'd need two arguments i.e. --subvol and --group
We already have the --subvol argument parsing and implementation available downstream.
We just need to pull that back in upstream.

1. fetching the list of subvol groups and
2. fetching the list of subvols for a particular subvol group
   we already have this interface, but we'll have to enhance it to look at the given subvol group

With no --group argument, the default subvol group will be assumed for the --subvol argument.

e.g.

ceph fs snap-schedule add <path> <schedule> <start-time> [--fs <fs-name>] [--subvol <subvol-name> [--group <subvol-group-name>]]

Note: We have now made the --fs argument mandatory for multi-fs scenario

An issue with snap-schedule implementation is that although the subvol name is recorded in the DB, the group name is never recorded. So we either have to alter the DB schema and hence deal with DB schema versions or we combine the subvol and group name in a single string and store that composite value in the subvol field and deal with insertion, extraction and display of those fields.

Milind Changire wrote:

An issue with snap-schedule implementation is that although the subvol name is recorded in the DB, the group name is never recorded. So we either have to alter the DB schema and hence deal with DB schema versions or we combine the subvol and group name in a single string and store that composite value in the subvol field and deal with insertion, extraction and display of those fields.

That would be confusing. Let's maintain it as a separate column in the db. Can the db schema be upgraded on the fly for older db versions?

Venky Shankar wrote:

Milind Changire wrote:

An issue with snap-schedule implementation is that although the subvol name is recorded in the DB, the group name is never recorded. So we either have to alter the DB schema and hence deal with DB schema versions or we combine the subvol and group name in a single string and store that composite value in the subvol field and deal with insertion, extraction and display of those fields.

That would be confusing. Let's maintain it as a separate column in the db. Can the db schema be upgraded on the fly for older db versions?

Introducing schema versions is not that difficult. It's automated for the mgr module default db (in the .mgr pool). See:

https://github.com/ceph/ceph/blob/f9626b596907707d03e7544681d3b58c8c30abd9/src/pybind/mgr/mgr_module.py#L1203-L1232

I think copy/paste as needed (no need to refactor I think, not worth the effort).

for snap-schedule, the --subvol and --group arguments are used only by the "fs snap-schedule status" and "fs snap-schedule add" commands
all other commands work with the absolute path
do I need to introduce the --subvol and --group arguments for the other commands as well ?

Milind Changire wrote:

for snap-schedule, the --subvol and --group arguments are used only by the "fs snap-schedule status" and "fs snap-schedule add" commands
all other commands work with the absolute path
do I need to introduce the --subvol and --group arguments for the other commands as well ?

Yes! If we are fixing it then may as well do a thorough fix.