Learning Ceph - Second Edition

PG states

Every PG is assigned a set of states at any given time. In a day-to-day operation, a healthy PG shows active+clean. The active state indicates that the PG has no problems serving client I/O and clean implies that it fulfills other requirements including distributing its copies across a sufficient number of devices as dictated by the size attribute of the pool.

An unhealthy or degraded PG is denoted by a change in its states; for example, it has either stopped serving I/O, has lost communication with peers, or has fewer than required copies of its Objects. Each PG's states can include one or more of those described here. Ceph endeavors to return unhealthy PGs back to their healthy states and rebalances data to fulfill the constraints ...