Monitoring Ceph placement groups
Ceph objects are mapped to distinct and unique placement groups (PGs) that are stored by OSDs. The health of the cluster depends on the state and health of the placement groups, both individually and collectively. If a placement group is degraded then the health of the cluster is impacted, even though the rest of the PGs may be healthy and do not experience issues with client I/O. The cluster status will be HEALTH_OK only when all placement groups are in active and clean states. An unhealthy cluster manifests PGs that may be either in the active or clean state but not both—and possibly neither. The active state is required to serve client I/O, while clean indicates that the PG is not only serving I/O but also meeting other necessary requirements including replication.
PG states
We can divide PG states into three categories: Critical, Warning, and Info. PG states in the Critical category indicate severe damage to the cluster and immediate action should be taken...
