[
https://issues.apache.org/jira/browse/OOZIE-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16519627#comment-16519627
]
Vinod Kumar Vavilapalli commented on OOZIE-3227:
------------------------------------------------
MAPREDUCE-4503 is the right Jira which introduced the incompatibility in trunk,
linking it. The incompatibility doesn't exist in Hadoop 2.x because of the
patch at MAPREDUCE-4549.
> Eliminate duplicate dependencies when using Hadoop 3 DistributedCache
> ---------------------------------------------------------------------
>
> Key: OOZIE-3227
> URL: https://issues.apache.org/jira/browse/OOZIE-3227
> Project: Oozie
> Issue Type: Sub-task
> Components: core
> Affects Versions: 5.0.0, 4.3.1
> Reporter: Denes Bodo
> Assignee: Denes Bodo
> Priority: Major
> Fix For: 5.1.0
>
> Attachments: OOZIE-3227_001.patch, OOZIE-3227_002.patch
>
>
> Using Hadoop 3 it is not allowed to have multiple dependencies with same file
> names on the list of *mapreduce.job.cache.files*.
> The issue occurs when I have the same file name on multiple sharelib folders
> and/or my application's lib folder. This can be avoided but not easy all the
> time.
> I suggest to remove the duplicates from this list.
> A quick workaround for the source code in JavaActionExecutor is like:
> {code}
> removeDuplicatedDependencies(launcherJobConf,
> "mapreduce.job.cache.files");
> removeDuplicatedDependencies(launcherJobConf,
> "mapreduce.job.cache.archives");
> ......
> private void removeDuplicatedDependencies(JobConf conf, String key) {
> final Map<String, String> nameToPath = new HashMap<>();
> StringBuilder uniqList = new StringBuilder();
> for(String dependency: conf.get(key).split(",")) {
> final String[] arr = dependency.split("/");
> final String dependencyName = arr[arr.length - 1];
> if(nameToPath.containsKey(dependencyName)) {
> LOG.warn(dependencyName + " [" + dependency + "] is already
> defined in " + key + ". Skipping...");
> } else {
> nameToPath.put(dependencyName, dependency);
> uniqList.append(dependency).append(",");
> }
> }
> uniqList.setLength(uniqList.length() - 1);
> conf.set(key, uniqList.toString());
> }
> {code}
> Other way is to eliminate the deprecated
> *org.apache.hadoop.filecache.DistributedCache*.
> I am going to have a deeper understanding how we should use distributed cache
> and all the comments are welcome.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
