When an OSD is marked out, the affected PGs will re-peer with new OSDs and start the process of recovering and backfilling data across the cluster. This process can put a strain on the disks in a Ceph cluster and lead to higher latencies for client requests. There are several tuning options that can reduce the impact of backfilling by reducing the rate and priority. These should be evaluated against the impact of slower recovery from failed disks, which may reduce the durability of the cluster.