October 2017
Intermediate to advanced
340 pages
9h 46m
English
When any component within a cluster fails, be it a single OSD device, a host's worth of OSDs, or a larger bucket like a rack, Ceph waits for a short grace period before it marks the failed OSDs out. This state is then updated in the CRUSH map. As soon an OSD is marked out, Ceph initiates recovery operations. This grace period before marking OSDs out is set by the optional ceph.conf tunable mon_osd_down_out_interval, which defaults to 300 seconds (5 minutes). During recovery Ceph moves or copies all data that was hosted on the OSD devices that failed.
Since CRUSH replicates data to multiple OSDs, replicated copies survive and are read during recovery. As CRUSH develops the requisite new mapping of PGs to ...
Read now
Unlock full access