Despite plenty of available mapper slots (as inferred by "Map Task Capacity" minus "Maps" on the jobtracker) our jobs often linger with all or many of the map tasks stuck in a pending status for five to ten minutes. Eventually they start, but why the seemingly unnecessary delay? There also might be a correlation between starting an unrelated job to "unstick" the pending tasks of the previous job...although this correlation is difficult to quantify; it may be a coincidence.
Does anyone have a theory as to what is going on here?
Keith Wiley [email protected] keithwiley.com music.keithwiley.com
"Luminous beings are we, not this crude matter."
Grokbase › Groups › Hadoop › common-dev › March 2011