On Aug 2, 2011, at 12:19 PM, <milind.bhandar...@emc.com> <milind.bhandar...@emc.com> wrote: > 1. Cut a release 0.22.0 without mapreduce-2178 patch, with > hadoop.security.authentication set to simple (I.e. No authentication). > Make sure that MR-2178 is highlighted as known-issue in the top-level dir > (create a KNOWN-ISSUES.txt ?), and carry out a vote.
This is a great way to end up on CERT, "known issue" or not. That's not a place where you want to be. > The reason a 0.22.0 release should happen, IMHO, even without MR-2178, is > that the other components of Hadoop stack (e.g. Pig, hbase, hive etc) will > have a release to test against, and fix any issues for incompatibility. > Also, for folks who do not use kerberos authentication (>90% of the hadoop > users, to my knowledge), I'm 99% certain that the task controller code is independent of Kerberos usage since it has its own conflg var... > it will be a release closest to trunk, with HDFS > appends. Plus, any issues found in the HDFS part of the release would be > directly applicable to the trunk and other future releases. (Not so about > map-reduce, which will undergo major surgery in trunk because of MR-279.) > > Thoughts ? I'd rather have the task-controller code ripped out than shipped broken.