I found the following to be a good discussion of the same topic: http://apache-spark-user-list.1001560.n3.nabble.com/The-concurrent-model-of-spark-job-stage-task-td13083.html
> From: so...@cloudera.com > Date: Tue, 20 Jan 2015 10:02:20 +0000 > Subject: Re: Does Spark automatically run different stages concurrently when > possible? > To: paliwalash...@gmail.com > CC: davidkl...@hotmail.com; user@spark.apache.org > > You can persist the RDD in (2) right after it is created. It will not > cause it to be persisted immediately, but rather the first time it is > materialized. If you persist after (3) is calculated, then it will be > re-calculated (and persisted) after (4) is calculated. > > On Tue, Jan 20, 2015 at 3:38 AM, Ashish <paliwalash...@gmail.com> wrote: > > Sean, > > > > A related question. When to persist the RDD after step 2 or after Step > > 3 (nothing would happen before step 3 I assume)? > > > > On Mon, Jan 19, 2015 at 5:17 PM, Sean Owen <so...@cloudera.com> wrote: > >> From the OP: > >> > >> (1) val lines = Import full dataset using sc.textFile > >> (2) val ABonly = Filter out all rows from "lines" that are not of type A > >> or B > >> (3) val processA = Process only the A rows from ABonly > >> (4) val processB = Process only the B rows from ABonly > >> > >> I assume that 3 and 4 are actions, or else nothing happens here at all. > >> > >> When 3 is invoked, it will compute 1, then 2, then 3. 4 will happen > >> after 3, and may even cause 1 and 2 to happen again if nothing is > >> persisted. > >> > >> You can invoke 3 and 4 in parallel on the driver if you like. That's > >> fine. But actions are blocking in the driver. > >> > >> > >> > >> On Mon, Jan 19, 2015 at 8:21 AM, davidkl <davidkl...@hotmail.com> wrote: > >>> Hi Jon, I am looking for an answer for a similar question in the doc now, > >>> so > >>> far no clue. > >>> > >>> I would need to know what is spark behaviour in a situation like the > >>> example > >>> you provided, but taking into account also that there are multiple > >>> partitions/workers. > >>> > >>> I could imagine it's possible that different spark workers are not > >>> synchronized in terms of waiting for each other to progress to the next > >>> step/stage for the partitions of data they get assigned, while I believe > >>> in > >>> streaming they would wait for the current batch to complete before they > >>> start working on a new one. > >>> > >>> In the code I am working on, I need to make sure a particular step is > >>> completed (in all workers, for all partitions) before next transformation > >>> is > >>> applied. > >>> > >>> Would be great if someone could clarify or point to these issues in the > >>> doc! > >>> :-) > >>> > >>> > >>> > >>> > >>> -- > >>> View this message in context: > >>> http://apache-spark-user-list.1001560.n3.nabble.com/Does-Spark-automatically-run-different-stages-concurrently-when-possible-tp21075p21227.html > >>> Sent from the Apache Spark User List mailing list archive at Nabble.com. > >>> > >>> --------------------------------------------------------------------- > >>> To unsubscribe, e-mail: user-unsubscr...@spark.apache.org > >>> For additional commands, e-mail: user-h...@spark.apache.org > >>> > >> > >> --------------------------------------------------------------------- > >> To unsubscribe, e-mail: user-unsubscr...@spark.apache.org > >> For additional commands, e-mail: user-h...@spark.apache.org > >> > > > > > > > > -- > > thanks > > ashish > > > > Blog: http://www.ashishpaliwal.com/blog > > My Photo Galleries: http://www.pbase.com/ashishpaliwal > > --------------------------------------------------------------------- > To unsubscribe, e-mail: user-unsubscr...@spark.apache.org > For additional commands, e-mail: user-h...@spark.apache.org >
