The number doesn't look bad.
At most, 160 x 7 = 1120 coordinator actions can compete for data dependency checks. Is it possible to reduce the throttle number (for example 3)? It should not create much problem. Also did you change other relevant configuration? something like "oozie.service.coord.input.check.requeue.interval"? Also, if possible , can you pls check the queue.size in oozie-instrumentation.log around the time when this problem happens? This value is updated every minutes. Regards, Mohammad ________________________________ From: Shangzhong zhu <[email protected]> To: [email protected]; Mohammad Islam <[email protected]> Sent: Friday, September 13, 2013 9:19 AM Subject: Re: oozie coordinator action start time delay Hi Mohammad, We have about 160 active coordinator jobs in the system, Active workflow jobs is around 15 at a given time. Throttle: 7 conCur: 1 Thanks. On Fri, Sep 13, 2013 at 12:11 AM, Mohammad Islam <[email protected]> wrote: > > > Hi Shangzhong, > How many jobs are active in the system? > > What are the values of "throttle" and "concurrency" in your coordinator > files? > More details : > > http://oozie.apache.org/docs/3.3.2/CoordinatorFunctionalSpec.html#a6.1.6._Coordinator_Action_Execution_Policies > > > Regards, > Mohammad > > ________________________________ > From: Shangzhong zhu <[email protected]> > To: [email protected] > Sent: Thursday, September 12, 2013 6:41 PM > Subject: oozie coordinator action start time delay > > > Hi, we are running Oozie Incubation 3.2.0. > > The coordinator job is data trigger based. What we found was, sometimes, > there seems to be a long delay between the time the trigger file is ready, > until the time the coordinator action was started. > > For example, > > The trigger file was created at 2013-09-11 17:07, > The coordinator action nominal time is 2013-09-11 15:00 > The coordinator start time is actually 2013-09-11 17:25 > > In another word, the action started 18 minutes after the trigger file is > ready. > > Any idea what is going on here? and how to reduce delay. > > BTW, this seems to be a random behavior. Is it related to the loads on > Oozie server? > > Thanks. >
