Thanks Sangjin for the work! I’ve tested our temporary fix on HBase UT failures (bd5af9c) and it looks good to me. I’m not an expert in HTrace or HDFS, but so far the fix works on our side. We may need to revert our current fix after HDFS-9187 is officially done, though. For now I’m +1 on the temp fix in YARN-2928 branch.
Li Lu On Oct 12, 2015, at 18:02, Sangjin Lee <[email protected]<mailto:[email protected]>> wrote: Hi folks, I have completed the rebase of YARN-2928 (this time cherry-picks really) to the trunk as of last Saturday. I resolved 10 merge conflicts most of which were minor. But I do want to call out a few of them, and would like you to review how I resolved those conflicts before I make the rebase official. I have just pushed this new branch ("*YARN-2928-rebase*") so you can take a look at it. I'll swap the branches once we're satisfied. The following are those commits to review. I called out those who might be best to review the merges. [3e3a8fe: Junping] Trunk added a new use (in TestContainerResourceIncreaseRPC <https://github.com/apache/hadoop/blob/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-common/src/test/java/org/apache/hadoop/yarn/TestContainerResourceIncreaseRPC.java#L99>) of a method (TestRPC.newContainerToken) in TestRPC which we moved from yarn-common to yarn-server-common. I copied that method in TestContainerResourceIncreaseRPC. We could reconsider whether we want to move TestRPC from yarn-common to yarn-server-common. I don't recall the details of the discussion, but was there a strong reason to move TestRPC out of yarn-common? If trunk keeps creating new uses of this class, it might be a problem. [d35d861: Naga] Trunk added a new RM event type (app updated: YARN-4044 <https://github.com/apache/hadoop/commit/a9aafad12b1d2f67e55e09a6fa261d61789c9d7e>). I applied the same changes and moved code to AbstractTimelineServicePublisher, TimelineServiceV1Publisher, and TimelineServiceV2Publisher respectively. Naga, could you please confirm if that new event is done right in the merge commit? [bd5af9c] It turns out HDFS-9080 broke the HBase mini-cluster, which in turn broke our HBase-based unit tests. This was caught by HDFS-9187 which has a patch. The patch is not entirely correct (causes NPEs), and I applied a fixed version of that patch to our branch to ensure our tests pass. Let me know if you are OK with that. I don't think we can wait until HDFS-9187 gets resolved. If you could take a look at these commits, and let me know +1/-1, I'll be able to take the next steps. Thanks everyone! Regards, Sangjin
