hadoop-common-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aaron Baff <Aaron.B...@telescope.tv>
Subject Worker JVM's not terminating quickly with JVM reuse
Date Thu, 02 Dec 2010 19:21:49 GMT
I've encountered an issue where I have JVM reuse turned on (via setting in the Job configuration,
not the JobTracker), and when I submit another job immediately after the first one finishes,
it takes several seconds before any Map tasks begin. In looking at an individual Task node,
it appears that all of the worker JVM's need to terminate before any new Map tasks from a
different Job can be started. I understand that the JVM reuse only applies within a single
Job, but is there a setting to have the JobTracker tell the individual TaskTrackers that the
Job has finished and so they need to terminate the worker JVM's? Or is there a setting to
reduce the time that the worker JVM's hang around before terminating?


  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message