[KUDU-1437] Negative values corruption for RLE-encoded INT32 columns

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.6.0, 0.7.0, 0.8.0
• Fix Version/s: 0.9.0
• Component/s: cfile
• Target Version/s: 0.9.0
• Code Review: http://gerrit.cloudera.org:8080/#/c/2954/
• Assignee: Todd Lipcon
• Reporter: Przemyslaw Maciolek
• Created: 04/May/16 17:45
• Updated: 04/May/16 23:38
• Resolved: 04/May/16 23:38

Description

Putting a sequence of negative values to INT32 RLE-encoded column corrupts the data after compaction.

The exact recipe to reproduce it is following:

1) Put a sequence of negative values in descending order to an INT32 RLE-encoded column (it's a primary key in my case, but maybe regular columns are affected as well)
2) Wait for compaction
3) See you data converted to -1's

A Gist showing an example of data corruption is available here: https://gist.github.com/pmaciolek/dfc6a95ff85145d23774cb5d1a9f5a4f (based on sample.cc provided with Kudu sources, note that -std=c++11 might need to be added to the compiler flags)