[ https://issues.apache.org/jira/browse/HADOOP-12979?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15218072#comment-15218072 ]
Steve Loughran commented on HADOOP-12979: ----------------------------------------- The problem is the branch take if there's no s3 buffer dir defined. the else clause is broken; it's looking for a config option which isn't there. {code} if (conf.get(BUFFER_DIR, null) != null) { lDirAlloc = new LocalDirAllocator(BUFFER_DIR); } else { lDirAlloc = new LocalDirAllocator("${hadoop.tmp.dir}/s3a"); // HERE } {code} The fix should be to set {{BUFFER_DIR}} to the full path desired, create the {{LocalDirAllocator(BUFFER_DIR)}} from the (possibly enhanced) config > IOE in S3a: ${hadoop.tmp.dir}/s3a not configured > ------------------------------------------------- > > Key: HADOOP-12979 > URL: https://issues.apache.org/jira/browse/HADOOP-12979 > Project: Hadoop Common > Issue Type: Bug > Affects Versions: 2.8.0 > Reporter: Steve Loughran > > Running some spark s3a tests trigger an NPE in Hadoop <=2/7l IOE in 2.8 > saying "${hadoop.tmp.dir}/s3a not configured". > That's correct: there is no configuration option on the conf called > "${hadoop.tmp.dir}/s3a ". There may be one called {{hadoop.tmp.dir}}, however. > Essentially s3a is sending the wrong config option down, if it can't find > {{fs.s3a.buffer.dir}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)