OSD replication factor applies to all pools
Closed this issue · 3 comments
We adjust the replication factor for OSDs up to 3 when adding systems to a MicroCloud.
We do this by specifying Pools: []string{"*"} which applies the replication factor to all pools. However, if these pools' sizes were manually modified, MicroCloud will not have this context and will overwrite the user configuration on each run of microcloud add, microcloud service add, and even microcloud init with an existing MicroCeph.
We will need to add a GET /1.0/pools-op to MicroCeph to fetch current pool sizes from ceph.
With this endpoint, we can inspect the current pool configuration and make inferences about whether we should change it:
-
on
microcloud init, if the total OSD count is fewer than 3, we will set all "managed" OSD pool sizes tomax(3, count(OSDs)). -
on
microcloud add, andmicrocloud service add, present a list of "unmanaged" OSD pools whose size is smaller (or larger) than the new OSD count. The user will have to select the pools they want to increase in replication size. "Managed" pools will automatically be updated.
A "managed" pool is one that MicroCloud sets up. So that means lxd_remote, lxd_cephfs, lxd_cephfs_meta, lxd_cephfs_data, and .mgr.
Can we use the LXD client to query for both the remote and remote-fs pools?
This way we don't have to query Ceph directly.
Even that isn't quite enough because we also need to make sure the replication size isn't already customized to a higher value. Otherwise we would reset them all back to 3 each time a node is added.