ceph-mds: cephfs filesystems does not support erasure coded pools

For filesystems to use erasure coded pools, the data pool needs to be modified slightly.

ceph osd pool set ec_pool allow_ec_overwrites true

Describe the solution you’d like This could be implemented in the deploy filesystem pools block of create_mds_filesystems.yml as:

- name: customize pool for partial writes
  command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} osd pool set {{ item.name }} allow_ec_overwrites true"
  with_items: "{{ cephfs_pools | unique }}"
  changed_when: false
  when: item.type | default("1") == '3' or item.type | default("1") == 'erasure'

Describe alternatives you’ve considered A explicit flag could be introduced into cephfs_pools.allow_ec_overwrites that can be used to toggle this behavior. I don’t know why it shouldn’t always follow cephfs_pools.type == erasure though.

Additional context This might need to be tested with a fix to #4768 as it is unclear whether the cephfs_pools.type is used correct.

[kwaegema]

this should work now?

[tmartinx]

I came across this issue while trying to figure out how to set allow_ec_overwrites: true to an erasure-coded pool. Doesn’t look that it’s available yet, I’m manually setting ec_overwrites for each pool after ceph-ansible, which is no big deal actually… =P