Failures appear related to
MAPREDUCE-1275. Will try again.
When i tried to test the patch, I realized that the test timeout on
MAPREDUCE-1365 is because of MAPREDUCE-1371.
nod Yes, you're right. I hadn't tested that. The test timeout wasn't my motivation, but the spurious failure in
MAPREDUCE-64 that would be easier to diagnose.
I would only set the fatalError value if it is not null, so that the earliest fault gets retained. A setFatalError() method could do this.
I don't see what you mean. Each tracker retains its cause of death; it's not shared between them and each tracker should only set this once. Are you suggesting making the error global and retaining only the first fault across all trackers?
Also, this may be an opportunity to give the MiniMRCluster and MinDFS cluster a common base class rather than continue to duplicate code.
Refactoring the Mini*Clusters is out of scope for this issue. This is just making the cause of test failures related to