And the combiner runs while fetching the outputs right? -----Original Message----- From: Arun C Murthy [mailto:a...@yahoo-inc.com] Sent: Monday, August 03, 2009 9:27 AM To: mapreduce-...@hadoop.apache.org Cc: common-dev@hadoop.apache.org Subject: Re: why reduce task can be scheduled before map tasks are 100% completed?
That check ensures sufficient #maps are completed before any of the reduces for the job are started. The reduces 'shuffle' map outputs from completed maps, but don't get into the 'reduce' phase until all map-outputs are copied over. Arun PS: Moving this to mapreduce-dev@ On Aug 2, 2009, at 5:02 AM, 我的Gmail邮箱 wrote: > Hi, everyone. > In class org.apache.hadoop.mapred.JobInProgress, there is a public > method: > scheduleReduces(), it will return true if "finishedMapTasks >= > completedMapsForReduceSlowstart" > and then the scheduler can schedule a new reduce task for a given > taskTracker. > > but as I konw, reduce can not be started unitl map is 100% > completed. Does > anyone can explain it? thanks a lot.
