Here's a patch which addresses the issue. Over on
HDFS-2920 (where this patch was originally posted) Eli had the following feedback:
!nameNodeHasResourcesAvailable implies "The NameNode has run out of resources" instead of the "NameNode is low on resources". It would be even better if the message was more specific (eg mentioned lack of inodes of disk space)
Yea, agree, but to do that would be a more invasive change, and more fundamental to to the way the NN currently checks resources, to return a status and a reason. Might if I punt that to another JIRA?