All the findings are attached over here: MAPREDUCE-4820

At this point I'd agree with Arun and close this one. We can track what's left over at MAPREDUCE-4820

Ok, that is news to me. Can we please close this and open a new jira then? Tx.

I think the problem is different, Roman would you please post the findings you send me.

Thx

Also, verified that the same fix is present in branch-2.0.4-alpha.

So, not sure why Oozie is still having problems... Roman, can you please re-check?

Here is what I see on branch-2 from 'git log':

commit 9aba3ebb2d455932981cc37fe8e3fa7a6ec4da82
Author: Alejandro Abdelnur <tucu@apache.org>
Date: Tue Dec 11 19:50:32 2012 +0000

MAPREDUCE-4549. Distributed cache conflicts breaks backwards compatability. (Robert Evans via tucu)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1420363 13f79535-47bb-0310-9956-ffa450edef68

Making this a blocker for 2.0.4-alpha as agreed. I will keep attaching logs/debugging data from Oozie runs to MAPREDUCE-4820

Do we want it on trunk?

For example if I pass in -libjar a/foo.jar -libjar b/foo.jar in 1.0 everything will work and both foo.jar files will be on the classpath (I don't know why you would do this, but it is possible). Under YARN only one of them will even be shipped in the distributed cache, so there is no way at all for both of them to be on the classpath.

With this patch it will output a warning message when a conflict occurs so people can try and fix the issue, but eventually we want to have the upper parts of the stack fix it, or we need to do what I was looking at doing initially and try to keep the colliding files separate by adding in support of sub directories in the distributed cache and using UUIDs.

Has this been committed to trunk? Can we close this?

I am fine with this being in branch-2. +1

Committed to branch-2 now, I'll wait till FRI noon to see if there are objections for committing this to trunk as well.

+1 overall. Here are the results of testing the latest attachment
http://issues.apache.org/jira/secure/attachment/12560334/MAPREDUCE-4549-trunk.patch
against trunk revision .

+1 @author. The patch does not contain any @author tags.

+1 tests included. The patch appears to include 1 new or modified test files.

+1 javac. The applied patch does not increase the total number of javac compiler warnings.

+1 javadoc. The javadoc tool did not generate any warning messages.

+1 eclipse:eclipse. The patch built with eclipse:eclipse.

+1 findbugs. The patch does not introduce any new Findbugs (version 1.3.9) warnings.

+1 release audit. The applied patch does not increase the total number of release audit warnings.

+1 core tests. The patch passed unit tests in hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common.

+1 contrib tests. The patch passed contrib unit tests.

Test results: https://builds.apache.org/job/PreCommit-MAPREDUCE-Build/3118//testReport/
Console output: https://builds.apache.org/job/PreCommit-MAPREDUCE-Build/3118//console

This message is automatically generated.

+1 pending jenkins.

Reopened and uploaded a patch for trunk

This JIRA has been lingering for a while. Alejandro, if you want this in branch-2 feel free to merge it in there and then resolve the JIRA, if not, I will just resolve the JIRA myself in a week or so.

Integrated in Hadoop-Hdfs-0.23-Build #347 (See https://builds.apache.org/job/Hadoop-Hdfs-0.23-Build/347/)
MAPREDUCE-4549. Distributed cache conflicts breaks backwards compatability (Robert Evans via tgraves) (Revision 1374407)

Result = SUCCESS
tgraves : http://svn.apache.org/viewcvs.cgi/?root=Apache-SVN&view=rev&rev=1374407
Files :

• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/CHANGES.txt
• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common/src/main/java/org/apache/hadoop/mapreduce/v2/util/MRApps.java
• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common/src/test/java/org/apache/hadoop/mapreduce/v2/util/TestMRApps.java

I committed this to 0.23.3, leaving jira open until we get agreement on branch-2

+1 for 0.23 change. I'm going to commit this to 0.23.

Arun, Alejandro - is there agreement on reverting MAPREDUCE-4503 from 2.0 and pulling this jira in there too?

That is fine with me. I realize that a lot of people are not going to use 0.23 as a bridge between 1.0 and 2.0 so if we want it to be deprecated in 2.0 too that is fine. If we revert MAPREDUCE-4503 this patch should apply cleanly to branch-2 also.

No worries, no need to be sorry at all. Thanks for the details.

IMO we should do the warning for Hadoop 2.x as well. Maybe in Hadoop 3.x converted to an exception.

Alejandro,

Sorry you are right I was way too vague about who I talked with. "oozie people" does not really say anything about who they are or exactly what we talked about. Thank you for calling me out on this it was my bad.

I spoke with Mohammad Islam and Virag Kothari. The extent of the conversation was primarily me clarifying the differences between 1.0 behavior of the distributed cache, the behavior of the cache in 2.0/0.23 prior to MAPREDUCE-4503, and its behavior post MAPREDUCE-4503. After that we talked about my proposed fix here, of turning the exception into a warning. They both were OK with that but wanted to perhaps look into having the error checking in 2.0 also happen as new items were added through the distributed cache APIs so that they could react to the issues on a per entry basis, instead of having the job submission fail. I thought this was reasonable and told them to file a JIRA against MAPREDUCE for it.

I am happy for any other feedback people may have on the issues. The call was mostly for clarification. I probably didn't even need to mention it here, but since I did here is the full disclosure.

Where was this discussed with the Oozie people? I'm not aware of this, I did not see any discussion in the oozie-dev@ alias.

Would you please summarize the issue, how it affects Oozie and what is exactly the impact of leaving it incompatible in Hadoop 2?

At first look I think we should keep backwards compatibility in Hadoop 2.

+1 (non-binding), lgtm.

This patch will only apply to branch-0.23

After talking to Arun and several Oozie people about I have decided that for branch-2 and trunk we will keep the same behavior as now, and for 0.23.3 we will change the exception into a warning. This will give oozie, and others time to deal with the incompatibility and not block 0.23.3 from being released.

Integrated in Hadoop-Hdfs-0.23-Build #344 (See https://builds.apache.org/job/Hadoop-Hdfs-0.23-Build/344/)
svn merge -c -1369197 Reverting: MAPREDUCE-4503 in branch-0.23 until MAPREDUCE-4549 can be addressed. (Revision 1372573)

Result = SUCCESS
bobby : http://svn.apache.org/viewcvs.cgi/?root=Apache-SVN&view=rev&rev=1372573
Files :

• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/CHANGES.txt
• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common/src/main/java/org/apache/hadoop/mapreduce/v2/util/MRApps.java
• /hadoop/common/branches/branch-0.23/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common/src/test/java/org/apache/hadoop/mapreduce/v2/util/TestMRApps.java