Hi,
When I ran a job (containing some hundreds of thousands tasks) over our
hadoop-0.19.2 cluster, I got OutOfMemoryError at JobTracker. Monitoring
memory usage at the JobTracker with Ganglia, it looks like that memory space
of the JobTracker is released every 24 hours, which is the default value of
mapred.jobtracker.retirejob.interval.
What happens if I set mapred.jobtracker.retirejob.interval to any less
values like 12 hours or even 0? Does it only control job retirement interval
from JobTrackers memory? Any ohter side effects can you suppose?
Thanks.
Manhee
