[ https://issues.apache.org/jira/browse/HDFS-3577?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13409701#comment-13409701 ]
Daryn Sharp commented on HDFS-3577: ----------------------------------- bq. The file size was 1MB in the test but the block size was only 1kB. Therefore, it created a lot of local files and failed with "java.net.SocketException: Too many open files". Does this mean there's a fd leak? Or at least a leak during the create request? If so, is the test at fault? > webHdfsFileSystem fails to read files with chunked transfer encoding > -------------------------------------------------------------------- > > Key: HDFS-3577 > URL: https://issues.apache.org/jira/browse/HDFS-3577 > Project: Hadoop HDFS > Issue Type: Bug > Components: hdfs client > Affects Versions: 2.0.0-alpha > Reporter: Alejandro Abdelnur > Assignee: Tsz Wo (Nicholas), SZE > Priority: Blocker > Attachments: h3577_20120705.patch, h3577_20120708.patch > > > If reading a file large enough for which the httpserver running > webhdfs/httpfs uses chunked transfer encoding (more than 24K in the case of > webhdfs), then the WebHdfsFileSystem client fails with an IOException with > message *Content-Length header is missing*. > It looks like WebHdfsFileSystem is delegating opening of the inputstream to > *ByteRangeInputStream.URLOpener* class, which checks for the *Content-Length* > header, but when using chunked transfer encoding the *Content-Length* header > is not present and the *URLOpener.openInputStream()* method thrown an > exception. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira