Object Recovery Basics

Explains object recovery modes.

HPE Ezmeral Data Fabric Object Store features the ability to recover objects. Recovery operations include cleaning partially written objects, purging non-current versions of unversioned objects, and cleaning incomplete multi-part objects. Recovery also claims space by deleting dangling delete markers and objects marked for purge. In addition, recovery aggregates bucket statistics.
NOTE Recovery can run only on the node where the master copy of OLT table's first tablet is hosted​. Recovery must be triggered using the alias of the node to which the bucket is assigned. Triggering recovery using a wrong node alias leads to a failure.
There are two recovery modes: mini and full.
Mini recovery mode
In a mini recovery, the system scans for buckets to recover every hour. The system picks up buckets that were created or modified in the last hour and examines them for any recovery to perform. The recovery deletes dangling delete markers.
Full recovery mode
In a full recovery, the system scans ALL buckets every week and examines them for any recovery to perform. Similar to the mini recovery, the system deletes all incomplete multipart uploads and dangling delete markers.

See mc admin recovery start and mc admin recovery stop to start and stop recovery.