Avi Shinnar created MAHOUT-1467:
-----------------------------------

             Summary: ClusterClassifier read/writePolicy leak file handles
                 Key: MAHOUT-1467
                 URL: https://issues.apache.org/jira/browse/MAHOUT-1467
             Project: Mahout
          Issue Type: Bug
          Components: Clustering
    Affects Versions: 0.9
            Reporter: Avi Shinnar


org.apache.mahout.clustering.classify.ClusterClassifier.readPolicy/writePolicy 
methods leak file handles.

This leak causes a serious problems in setups that reuse JVMs for multiple 
tasks (as with "mapred.job.reuse.jvm.num.tasks").

In more detail:
The org.apache.mahout.clustering.classify.ClusterClassifier class has  static 
methods to read/write policies: readPolicy and writePolicy

Both of them open a SequenceFile.Reader/Writer and read/write from/to it, and 
then neglect to close it.

suggested fix:
add a call to reader.close() before the return statement in readPolicy
and
add a call to writer.close() before the return statement in writePolicy





--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to