[HDFS-9655] NN should start JVM pause monitor before loading fsimage

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 2.8.0, 3.0.0-alpha1
• Labels: supportability
• Assignee: John Zhuge
• Reporter: John Zhuge
• Created: 16/Jan/16 17:51
• Updated: 30/Aug/16 01:20
• Resolved: 20/Jan/16 22:49

Description

We have seen many cases of NameNode startup either extremely slow or even hung. Most of them were caused by insufficient heap size with regard to the metadata size. Those cases were resolved by increasing the heap size.

However it did take support team some time to root cause. JVM pause warning messages would greatly assist in such diagnosis, but NN starts JVM pause monitor after fsimage/edits loading.

Propose to start JVM pause monitor before loading fsimage/edits.