Hi all, I backported SnapshotInputFormat to hbase 0.94. This resulted in a huge performance gain while the scan is running. Here is the corresponding jira https://issues.apache.org/jira/browse/HBASE-8369. The overall performance gain is mitigated by a very long initialization period. It can take up to 30 minutes after the snapshot completes for the MR job to actually begin.
Does anybody have experience running SnapshotInputFormay on 0.94 or 0.98? Are you seeing the same long initialization period? Thanks! Dave
