[ https://issues.apache.org/jira/browse/MAPREDUCE-5918?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14091228#comment-14091228 ]
Ravi Prakash commented on MAPREDUCE-5918: ----------------------------------------- Hi Sergey! Thanks for your contribution! :-) This is a good find. Sorry for taking so long to come around to it. # Should we do this for the mapreduce api as well (org.apache.hadoop.mapreduce.lib.input) ? # To me it seems the problem really is that CodecPool is not checking if the instance of Codec being "returned" is not already in the pool. Would you agree? I agree that we should also set decompressor to null after we have returned it. > LineRecordReader can return the same decompressor to CodecPool multiple times > ----------------------------------------------------------------------------- > > Key: MAPREDUCE-5918 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-5918 > Project: Hadoop Map/Reduce > Issue Type: Bug > Affects Versions: 2.3.0 > Reporter: Sergey Murylev > Assignee: Sergey Murylev > Priority: Minor > Fix For: trunk > > Attachments: MAPREDUCE-5918.1.patch > > > LineRecordReader can return the same decompressor to CodecPool multiple times > if method close() called multiple times. In this case CodecPool doesn't > guarantee that it always return different decompressors. This issue can cause > some difficult reproducible and difficult diagnosable bugs in Hadoop based > programs. -- This message was sent by Atlassian JIRA (v6.2#6252)