[ https://issues.apache.org/jira/browse/HDFS-2964?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Harsh J updated HDFS-2964: -------------------------- Attachment: HDFS-2964-2.patch +1 on this. I cleaned up the message to be more informative (print path and bytes being compared), and updated the patch for trunk. Awaiting jenkins run. > No notice in any logs if dfs.datanode.du.reserved is greater than available > disk space > -------------------------------------------------------------------------------------- > > Key: HDFS-2964 > URL: https://issues.apache.org/jira/browse/HDFS-2964 > Project: Hadoop HDFS > Issue Type: Bug > Components: datanode > Affects Versions: 2.6.0 > Reporter: Robert J Berger > Priority: Minor > Attachments: HDFS-2964-1.patch, HDFS-2964-2.patch, HDFS-2964.patch > > > We spent a long time tracking down why a test hdfs cluster seemed to be > running fine, but would not allow the mapred system to come up complaining > that "could only be replicated to 0 nodes, instead of 1". > There were no namenode or datanode errors in any of the logs. hadoop fsck > said everything was good. At first glance dfsadmin -report looked good. It > wasn't until I realized that there was 0 Capacity available that we poked > around and found > https://groups.google.com/a/cloudera.org/group/scm-users/msg/a4252d6623adbc2d > which mentioned that the "reserverd space" might be greater than the disk > space available. And we did find that our dfs.datanode.du.reserved was indeed > higher than our actual since we were only testing a small cluster. > It seems that there should be some warning or error in the logs that say that. -- This message was sent by Atlassian JIRA (v6.3.4#6332)