tl;dr after we get hadoop2 build fixed, how about making hadoop2 the default for trunk precommit and trunk jenkins jobs?
Over in HBASE-6891, we've fixed the unit tests with hadoop2 on trunk that always fail or always hang. We are getting close on getting the hadoop2 + trunk build passing (right now there are a combination of tests that hang sometimes or fail sometimes -- and I've seen a run where the hbase-server module passed on hadoop2). Initially I was thinking that we'd create a two-headed precommit script that ran unit tests against trunk on hadoop1 and hadoop2. This was far down in a thread so may not have been considered by many folks. I'd like to suggest (actually Elliot's idea) that we set the default precommit trunk build to be against hadoop2. Over the "normal" jenkins runs, we'd move trunk to hadoop2 as well. We would however keep the 0.95 branch against hadoop 1.x and would thus get coverage of much of the new code under both setups. This buys us two things: 1) We get coverage against hadoop2, which will likely be the post singularity hbase should use as a default. 2) We end up having the same number of build jobs and avoid having two separate hadoop2 hadoop1 precommit builds for trunk. Thoughts? Jon -- // Jonathan Hsieh (shay) // Software Engineer, Cloudera // [email protected]
