Yes I've noticed this one and its related cousin, but not sure this is the same issue there; our job "properly" ends after 6 attempts. We'll try with disabled speculative mode anyway!
On 25 May 2016 at 00:13, Ted Yu <yuzhih...@gmail.com> wrote: > Have you taken a look at SPARK-14915 ? > > On Tue, May 24, 2016 at 1:00 PM, Adrien Mogenet < > adrien.moge...@contentsquare.com> wrote: > >> Hi, >> >> I'm wondering how Spark is setting the "index" of task? >> I'm asking this question because we have a job that constantly fails at >> task index = 421. >> >> When increasing number of partitions, this then fails at index=4421. >> Increase it a little bit more, now it's 24421. >> >> Our job is as simple as "(1) read json -> (2) group-by sesion identifier >> -> (3) write parquet files" and always fails somewhere at step (3) with a >> CommitDeniedException. We've identified that some troubles are basically >> due to uneven data repartition right after step (2), and now try to go >> further in our understanding on how does Spark behaves. >> >> We're using Spark 1.5.2, scala 2.11, on top of hadoop 2.6.0 >> >> -- >> >> *Adrien Mogenet* >> Head of Backend/Infrastructure >> adrien.moge...@contentsquare.com >> http://www.contentsquare.com >> 50, avenue Montaigne - 75008 Paris >> > > -- *Adrien Mogenet* Head of Backend/Infrastructure adrien.moge...@contentsquare.com http://www.contentsquare.com 50, avenue Montaigne - 75008 Paris