This in the context of providing backup/restore operation for SolrCloud (on HDFS). The copy operation is unavoidable in case Solr is using local filesystem and backup is being written on a shared file-system. But in case of HDFS, I am wondering if we can avoid the copy operation by using IndexDeletionPolicy.
On Thu, Apr 7, 2016 at 10:10 PM, Hrishikesh Gadre <[email protected]> wrote: > Hello, > > IndexDeletionPolicy in Lucene enables users to configure when the stale > Index commits should be deleted. Currently Solr uses this support to > "reserve" a commit point while the relevant segment files are being copied > as part of the backup operation. This reservation information is stored > in-memory only (and hence doesn't preserve across server restart). > > I wonder if we can use this to implement point-in-time and zero-copy > snapshots ?Essentially the idea would be store the IndexCommit related > information in a persistent fashion so that this "reservation" can be > preserved across server restart. Hence once the snapshot is created, the > relevant commit_point would be reserved permanently (unless the snapshot is > deleted). The advantage of this would be that we can eliminate the copy > operation. > > Any thoughts? > > Thanks > Hrishikesh > > > >
