[HBASE-2302] Optimize M-R by bulk excluding regions - less InputSplit-s to avoid traffic on region servers when performing M-R on a subset of the table

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.20.3
• Fix Version/s: 0.90.0
• Hadoop Flags: Reviewed
• Assignee: Unassigned
• Reporter: Karthik K
• Created: 09/Mar/10 03:51
• Updated: 20/Nov/15 12:42
• Resolved: 11/Mar/10 23:55

Description

TableInputFormatBase , creates a InputSplit per region. Given that the keys are sorted - sometimes - it might be needed to perform M-R on a subset of the keyset ( regions ) . Adding a provision to filter the regions when generating InputSplits might be useful .

The granularity of exclusion is per-region-wise. A RowFilter might still be needed during a Scan on a separate region, but that is a separate issue altogether.

Methodology: Add a way to prune the keyset before generating inputsplits , by default - all sets are returned indicating all regions are included that could be overridden as necessary, depending on the higher-level logic.