[ https://issues.apache.org/jira/browse/MAPREDUCE-1921?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12895096#action_12895096 ]
Dick King commented on MAPREDUCE-1921: -------------------------------------- {{MapTask.java}} When creating a new exception to re-throw, perhaps we should duplicate the original exception's type? I realize that there's a lot of code that doesn't do that, but swallowing this information feels wrong. Especially in the case of IO exceptions, I can envision a lot of code that wants to treat certain sub-exceptions a bit differently; some IO exceptions are the caller's fault but some represent oddities that happened in the file system. Perhaps something like {noformat} throw (IOException)(ioe.getClass().getConstructor(String.class, ioe.getClass()) .newInstance("IO error ..." + ... , ioe)); {noformat} ? This cliche should occur in two places; perhaps we should pull it out and put it in one of the utilities classes? Perhaps IO utilities, since the main use case where I can see people caring is the large variety of {{IOException}} s. > IOExceptions should contain the filename of the broken input files > ------------------------------------------------------------------ > > Key: MAPREDUCE-1921 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-1921 > Project: Hadoop Map/Reduce > Issue Type: Improvement > Reporter: Krishna Ramachandran > Assignee: Krishna Ramachandran > Attachments: mapred-1921-1.patch, mapred-1921-3.patch, > mapreduce-1921.patch > > > If bzip or other decompression fails, the IOException does not contain the > name of the broken file that caused the exception. > It would be nice if such actions could be avoided in the future by having the > name of the files that are broken spelled > out in the exception. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.