I've seen this before with JavaRDD. Some of our jobs perform much worse when we persist them in any form!
------- Regards, Andy On Tue, Mar 15, 2016 at 2:22 PM, Jakub Liska <liska.ja...@gmail.com> wrote: > Hey, > > I submitted this issue to Spark > https://issues.apache.org/jira/browse/SPARK-13909 > > But I think it is rather a Zeppelin issue because it came with upgrade to > current Zeppelin master version. It seems to be happening even on Spark > 1.5.x ... > > Any idea what might be wrong? Simply any kind of DISK_ONLY leads to OOME > ... I have a really simple single-node setup. >