Unless there is objection, in the next day or so, I'd like to revert the hadoop that is in hbase trunk and replace it with hadoop 0.20.2. Currently hbase trunk bundles the head of the 0.21 hadoop branch and will only run atop an hadoop built out of this hadoop 0.21 branch.
Since the plan is that the next release of hbase will run on hadoop 0.20.x and the major hadoop release that follows hadoop 0.20 -- 0.21 or 0.22, its version is still TBD -- AND given that the next major hadoop release would seem to be a few months out, having our trunk bundle hadoop 0.21 unnecessarily constrains where current hbase trunk can run. Even though the bundled hadoop 0.21 has the critical hdfs-265 working flush/sync feature, its looking like hard work by the fellas who are on the hdfs-200, hdfs-862 and hdfs-142 patchset will likely get us data durability on a (patched) hadoop 0.20.x some time soon. Having hbase trunk so it can work on the patched hadoop 0.20.x will help the testing of the data durability effort. Jean-Daniel, Ryan, and I have already had a pow-wow about this and think this is the way to go for now (The revert notion was originally Ryans -- See HBASE-2255). If you've been running hbase TRUNK and an hadoop 0.21, unfortunately, this means that you'll need to copy your data out of your cluster, do the downgrade, and then copy your data back in. If you need help with this effort, give us a shout and we'll help you out. Thanks, St.Ack
