Ceph » RADOS

Hello,
We faced with immediate object promotion when call lock on object.
This behavior makes very hard to understand workload and tune cache pool.
We use libradosstriper-based clients, so every read/write operation with stripped object causes immediate promotion of first part of stripped object.
If we work with a lot of small objects, we get a high network and disk i/o load. It highly reduces cache efficiency.

Ideally this behavior should be configurable.

It is possible to test with this command:

rados -p poolname lock get --lock-tag weird_ceph_locks --lock-cookie `uuid` objectname striper.lock

After running that command, object could be found in a cache pool.