-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/18638/
-----------------------------------------------------------
Review request for accumulo and Eric Newton.
Bugs: ACCUMULO-2422
https://issues.apache.org/jira/browse/ACCUMULO-2422
Repository: accumulo
Description
-------
Refinement and expansion of the initial fix for ACCUMULO-2242. The primary
improvement is skipping the renewal of the watch on another master's node after
a master manages to acquire the lock; the watch is no longer useless, and was
causing an error in the master log.
A lot of trace-level logging was added too, to help troubleshooters figure
things out.
Note that the diff is compared to ecn's initial fix.
Diffs
-----
fate/src/main/java/org/apache/accumulo/fate/zookeeper/ZooLock.java
25f1020be5f7189d6d7b40b57a35e059d1e298e7
server/src/main/java/org/apache/accumulo/server/master/Master.java
039d7fa9937409ab09ed0764fd87642001fdde51
Diff: https://reviews.apache.org/r/18638/diff/
Testing
-------
Easily created deadlock scenario by inserting a delay between when a master
gets a lock and when it updates its node's data. Tested on cluster with 2
masters. Tested fixed code and observed correct watcher renewals and assumption
of active master duties.
Thanks,
Bill Havanki