[HADOOP-5890] Use exponential backoff on Thread.sleep during DN shutdown

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.21.0
• Assignee: Todd Lipcon
• Reporter: Todd Lipcon
• Created: 22/May/09 00:29
• Updated: 24/Aug/10 20:38
• Resolved: 25/May/09 09:43

Description

Tests waste a lot of time in DataNode.shutdown. Typical logs look like:

2009-05-21 17:13:20,177 INFO  datanode.DataNode (DataNode.java:shutdown(637)) - Waiting for threadgroup to exit, active threads is 0
2009-05-21 17:13:20,177 INFO  datanode.DataBlockScanner (DataBlockScanner.java:run(620)) - Exiting DataBlockScanner thread.
2009-05-21 17:13:21,117 INFO  datanode.DataNode (DataNode.java:shutdown(637)) - Waiting for threadgroup to exit, active threads is 0

In this example (and very commonly) the DataBlockScanner thread exits within 5-10ms after the first wait. The DN then sleeps an entire second before succeeding in shutting down.

Using exponential backoff from a short value like 2ms up to a maximum of 1000ms would solve this.