On Wed, Oct 21, 2015 at 10:01 PM, 陈恒 <chenh...@fenbi.com> wrote: > When i dig in HBASE-14659, i found in testcase TestHFileOutputFormat, > some tests run MR job > without MiniMRCluster and some run with it. > > > The ones with no MiniMRCluster run much faster! > Do we need to 'test' clustered MR especially in a unit test? > > > We have some fat unit tests that spin up loads of mini clusters. For MR, the same code base is exercised whether cluster or not so if no MR cluster means less resources used by test and all runs faster, I'd be +1 in favor of undoing cluster (for all but the cases where cluster needed to prove some code as per example Elliott raises above).
Thanks 陈恒 St.Ack