[HBASE-19317] Increase "yarn.nodemanager.disk-health-checker.max-disk-utilization-per-disk-percentage" to avoid host-related failures on MiniMRCluster

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.0.0-beta-1, 2.0.0
• Component/s: integration tests, test
• Hadoop Flags: Reviewed
• Assignee: Josh Elser
• Reporter: Josh Elser
• Created: 21/Nov/17 18:39
• Updated: 01/Aug/18 06:21
• Resolved: 22/Nov/17 22:32

Description

YARN (2.7.4, at least) defaults to asserting at least 10% of the disk usage free on the local machine in order for the NodeManagers to function.

On my development machine, despite having over 50G free, I would see the warning from the NM that all the local dirs were bad which would cause the test to become stuck waiting to submit a mapreduce job. Surefire would eventually kill the process.

We should increase this value to avoid it causing us headache.