Oozie job expiring on Java action when writing to HDFS

Question

I have an Oozie coordinator that runs a workflow every hour. The workflow is composed of two sequential actions: a shell action and a Java action. When I run the coordinator, the shell action seems to execute successfully, however, when it's time for the Java action, the Job Browser in Hue always show:

There was a problem communicating with the server: Job application_<java-action-id> has expired.

When I click on the application_id, here's the snapshot:

This seems to point on views.py and api.py. When I looked into server logs:

[23/Nov/2015 02:25:22 -0800] middleware   INFO     Processing exception: Job application_1448245438537_0010 has expired.: Traceback (most recent call last):
  File "/usr/lib/hue/build/env/lib/python2.6/site-packages/Django-1.6.10-py2.6.egg/django/core/handlers/base.py", line 112, in get_response
    response = wrapped_callback(request, *callback_args, **callback_kwargs)
  File "/usr/lib/hue/build/env/lib/python2.6/site-packages/Django-1.6.10-py2.6.egg/django/db/transaction.py", line 371, in inner
    return func(*args, **kwargs)
  File "/usr/lib/hue/apps/jobbrowser/src/jobbrowser/views.py", line 67, in decorate
    raise PopupException(_('Job %s has expired.') % jobid, detail=_('Cannot be found on the History Server.'))
PopupException: Job application_1448245438537_0010 has expired.

The Java action consists of two parts: REST API call and writing to HDFS (via Hadoop client library) the parsed result. Eventhough the Java action job is expiring / failing on Job Browser, the write to HDFS was successful. Here's the snippet of the HDFS writing part of the Java code.

FileSystem hdfs = FileSystem.get(new URI(hdfsUriPath), conf);
OutputStream os = hdfs.create(file);
BufferedWriter br = new BufferedWriter(new OutputStreamWriter(os, "UTF-8"));
...
br.write(toWriteToHDFS);
br.flush();
br.close();
hdfs.close();

When I run the workflow as a standalone, I've got a 50-50 chance of success and expiration on the Java action part, but on coordinator, all Java action's are expiring.

The YARN logs shows this:

 Job commit failed: java.io.IOException: Filesystem closed
at org.apache.hadoop.hdfs.DFSClient.checkOpen(DFSClient.java:794)
at org.apache.hadoop.hdfs.DFSClient.create(DFSClient.java:1645)
at org.apache.hadoop.hdfs.DFSClient.create(DFSClient.java:1587)
at org.apache.hadoop.hdfs.DistributedFileSystem$6.doCall(DistributedFileSystem.java:397)
at org.apache.hadoop.hdfs.DistributedFileSystem$6.doCall(DistributedFileSystem.java:393)
at org.apache.hadoop.fs.FileSystemLinkResolver.resolve(FileSystemLinkResolver.java:81)
at org.apache.hadoop.hdfs.DistributedFileSystem.create(DistributedFileSystem.java:393)
at org.apache.hadoop.hdfs.DistributedFileSystem.create(DistributedFileSystem.java:337)
at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:908)
at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:889)
at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:786)
at org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler$EventProcessor.touchz(CommitterEventHandler.java:265)
at org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler$EventProcessor.handleJobCommit(CommitterEventHandler.java:271)
at org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler$EventProcessor.run(CommitterEventHandler.java:237)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:745) 

So it looks like it has problem with closing the FileSystem at the ending of my Java code (should I keep the FileSystem open?).

I'm using Cloudera Quickstart CDH 5.4.0 and Oozie 4.1.0

Answer 1

The problem is already solved. My Java action uses an instance (say variable fs) of org.apache.hadoop.fs.FileSystem class. At the end of the Java action, I use fs.close(), which will cause the problem on the next period of Oozie job. So when I removed this line, everything went well again.