As Ceph stripes data across all disks in the cluster, a single disk, which is in the process of failing but has not yet completely failed, may start to cause slow or blocked IO across the cluster. Often, this will be caused by a disk that is suffering from a large number of read errors, but it is not severe enough for the disk to completely fail. Normally, a disk will only reallocate sectors when a bad sector is written to. Monitoring the SMART stats from the disks will normally pick up conditions such as these and allow you to take action.