[GORA-130] gora-accumulo caches tablet locations between map reduce jobs

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.2
• Fix Version/s: 0.2.1
• Component/s: gora-accumulo
• Assignee: Keith Turner
• Reporter: Keith Turner
• Created: 11/May/12 14:45
• Updated: 01/Jul/13 05:05
• Resolved: 11/May/12 15:05

Description

Enis added a new Loop program to goraci that continually runs Generation and Verification map reduce jobs. So you have one process launching multiple map reduce jobs. I was running this I noticed an issue. After the first round of generation, the table had 16 tablets. So verification ran with 16 mappers, one per tablet. Then more data was inserted and the table split to 32 tablets. When verification ran again it started 16 mappers instead of 32. Turns out the gora-accumulo store was using stale cached information about the table to create the input splits for the map reduce job.

This issues will not affect the simple usage pattern of a single java process launching one map reduce job that reads from accumulo.