On 0.20 head, the indexed contrib tests fail intermittently -- a test deadlocks
and/or fails to terminate, leading to a direct test failure or the failure of
the next to initialize. I thought this might be HBASE-2322. Likelihood of
failure depends on which server I run the test, definitely a race.
Somewhere in the change set between 0.20_pre_durability and 0.20 branch head is
a difference that allows the tests to pass. (If it were a consistent failure I
could easily git bisect to find the commit.) If I stitch 0.20 branch together
leaving out group commit and other write path changes also I can seemingly
avoid the problem.
The patch on HBASE-2248 trips up against indexed contrib in a different way.
But I wonder if it is the test or even the contrib.
- Andy
