It is indeed a bug. I believe the shutdown procedure in #7820 only kicks in when the external shuffle service is enabled (a pre-requisite of dynamic allocation). As a workaround you can use dynamic allocation (you can set spark.dynamicAllocation.maxExecutors and spark.dynamicAllocation.minExecutors to the same value.
I'll file a Jira ticket. On Wed, Oct 7, 2015 at 10:14 AM, Alexei Bakanov <russ...@gmail.com> wrote: > Hi > > I'm running Spark 1.5.1 on Mesos in coarse-grained mode. No dynamic > allocation or shuffle service. I see that there are two types of temporary > files under /tmp folder associated with every executor: /tmp/spark-<UUID> > and /tmp/blockmgr-<UUID>. When job is finished /tmp/spark-<UUID> is gone, > but blockmgr directory is left with all gigabytes in it. In Spark 1.4.1 > blockmgr-<UUID> folder was under /tmp/spark-<UUID>, so when /tmp/spark > folder was removed blockmgr was gone too. > Is it a bug in 1.5.1? > > By the way, in fine-grain mode /tmp/spark-<UUID> folder does not get > removed in neither 1.4.1 nor 1.5.1 for some reason. > > Thanks, > Alexei > -- -- Iulian Dragos ------ Reactive Apps on the JVM www.typesafe.com