The first thing to do is have the Cluster Manager remove all excess buckets. That leaves you with fewer to examine/back up. You may have noticed that some bucket names begin with "db_" and some begin with "rb_". The former designates a primary bucket and the latter is for replica buckets. So you might think you can just back up the db_* files and avoid duplicates. Yes, you can, HOWEVER, if a primary bucket is lost one of the replicas will be made the new primary without changing the name. That means skipping the rb_* files could keep a primary bucket from being backed up. The likelihood of that happening depends on how often you back up the cluster and how often an indexer fails. To find the true primary buckets, ask the CM using this REST API call | rest /services/cluster/master/buckets search=has_primary=true Then use the output to drive the backups.
... View more