Hi Sendoh, Flink does not necessarily spread the tasks across all available TMs. It depends a little bit on the topology of the job, because in general it is beneficial to deploy downstream tasks to the same machines where their input tasks run. I described the process in more detail here [1]. This link [2] also contains some information regarding Flink's internal scheduling.
[1] http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/Scheduling-task-slots-in-round-robin-tp12068p12143.html [2] https://ci.apache.org/projects/flink/flink-docs-release-1.2/internals/job_scheduling.html Cheers, Till On Tue, May 23, 2017 at 6:03 PM, Sendoh <unicorn.bana...@gmail.com> wrote: > Hi Flink users, > > I saw our Flink cluster doesn't dispatch jobs equally. > So some nodes are busy and have few task slots left, and some nodes are not > busy and have many task slots. > > We don't have yarn setup. Would that be the reason? > > Or does Job manager dispatch job first to some nodes are not busy(more > memory left and task slots)? or Job manager doesn't consider it? > > Best, > > Sendoh > > > > -- > View this message in context: http://apache-flink-user- > mailing-list-archive.2336050.n4.nabble.com/How-does-Job- > manager-schedule-jobs-to-task-managers-tp13274.html > Sent from the Apache Flink User Mailing List archive. mailing list archive > at Nabble.com. >
