The uffizi content storage is getting dangerously close to full.
After discussing our requirements (redundancy, migration to ceph, archiver issues), the following plan is being put into place:
- reduce archiver lag and improve redundancy : Synchronously write all objects to azure as well as local storage
- reduce local storage requirements : Stripe the local storage of contents across uffizi and banco
- prepare the ceph migration, improve redundancy : Synchronously write a copy of the objects in ceph, with the final redundancy requirements (erasure coding, 5 chunks + 2 parity chunks), but with a more risky failure domain (OSD instead of host).
- reclaim storage space : remove already backed up objects from uffizi and banco
- future-proof the system : adapt the archiver to handle striped storages