[HADOOP-4374] JVM should not be killed but given an opportunity to exit gracefully

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.19.0
• Fix Version/s: 0.20.0

Release Note

This patch (1) Adds a shutdownHook that does syncLogs sothat logs of the current task are flushed and log.index is up to date in cases like System.exit(), or killed using signals(other than SIGKILL).
(2) Changes writeToIndexFile() to write to a temporary index file first and then rename to log.index sothat updates to log.index file are atomic.

• Assignee: Ravi Gummadi
• Reporter: Devaraj Das
• Created: 08/Oct/08 12:20
• Updated: 08/Jul/09 16:53
• Resolved: 01/Apr/09 05:21

Description

When the tasktracker picks an idle JVM for purging, it should signal the JVM to exit gracefully, rather than forcefully killing it. This might have the unfortunate side effect of logs not fully flushed yet in some cases.