GC limit overhead exceeded is usually sign of either inadequate heap size (not the case here) or application produces garbage (temp objects) faster than garbage collector collects them - GC consumes most CPU cycles. 17G of Java heap is quite large for many application and is above "safe and recommended" limit (6-8GB) for Java server application.
>From what I saw in the stack trace I can conclude that some operations in RDD implementation are heap polluters. I am not the expert in Spark but it seems that Spark is not well optimized yet to work with reasonably large Java heaps. One of the options here is try to reduce JVM heap size and reduce data size per JVM instance. -Vladimir Rodionov On Thu, Oct 30, 2014 at 5:22 AM, Ilya Ganelin <ilgan...@gmail.com> wrote: > The split is something like 30 million into 2 milion partitions. The > reason that it becomes tractable is that after I perform the Cartesian on > the split data and operate on it I don't keep the full results - I actually > only keep a tiny fraction of that generated dataset - making the overall > dataset tractable ( I neglected to mention this in the first email). > > The way the code is structured I have forced linear execution until this > point so at the time of execution of the split it is the only thing > happening. In terms of memory I have assigned 23gb of memory and 17gb of > heap. > On Oct 30, 2014 3:32 AM, "Sean Owen" <so...@cloudera.com> wrote: > >> Can you be more specific about numbers? >> I am not sure that splitting helps so much in the end, in that it has >> the same effect as executing a smaller number at a time of the large >> number of tasks that the full cartesian join would generate. >> The full join is probably intractable no matter what in this case? >> The OOM is not necessarily directly related. It depends on where it >> happened, what else you are doing, how much memory you gave, etc. >> >> On Thu, Oct 30, 2014 at 3:29 AM, Ganelin, Ilya >> <ilya.gane...@capitalone.com> wrote: >> > Hey all – not writing to necessarily get a fix but more to get an >> > understanding of what’s going on internally here. >> > >> > I wish to take a cross-product of two very large RDDs (using >> cartesian), the >> > product of which is well in excess of what can be stored on disk . >> Clearly >> > that is intractable, thus my solution is to do things in batches - >> > essentially I can take the cross product of a small piece of the first >> data >> > set with the entirety of the other. To do this, I calculate how many >> items >> > can fit into 1 gig of memory. Next, I use RDD.random Split() to >> partition >> > the first data set. The issue is that I am trying to partition an RDD of >> > several million items into several million partitions. This throws the >> > following error: >> > >> > I would like to understand the internals of what’s going on here so >> that I >> > can adjust my approach accordingly. Thanks in advance. >> > >> > >> > 14/10/29 22:17:44 ERROR ActorSystemImpl: Uncaught fatal error from >> thread >> > [sparkDriver-akka.actor.default-dispatcher-16] shutting down ActorSystem >> > [sparkDriver] >> > java.lang.OutOfMemoryError: GC overhead limit exceeded >> > at com.google.protobuf_spark.ByteString.toByteArray(ByteString.java:213) >> > at >> akka.remote.MessageSerializer$.deserialize(MessageSerializer.scala:24) >> > at >> > >> akka.remote.DefaultMessageDispatcher.payload$lzycompute$1(Endpoint.scala:55) >> > at akka.remote.DefaultMessageDispatcher.payload$1(Endpoint.scala:55) >> > at akka.remote.DefaultMessageDispatcher.dispatch(Endpoint.scala:73) >> > at >> > >> akka.remote.EndpointReader$$anonfun$receive$2.applyOrElse(Endpoint.scala:764) >> > at akka.actor.ActorCell.receiveMessage(ActorCell.scala:498) >> > at akka.actor.ActorCell.invoke(ActorCell.scala:456) >> > at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:237) >> > at akka.dispatch.Mailbox.run(Mailbox.scala:219) >> > at >> > >> akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:386) >> > at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260) >> > at >> > >> scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339) >> > at >> scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979) >> > at >> > >> scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107) >> > Exception in thread "main" java.lang.OutOfMemoryError: GC overhead limit >> > exceeded >> > at java.util.Arrays.copyOfRange(Arrays.java:2694) >> > at java.lang.String.<init>(String.java:203) >> > at java.lang.String.substring(String.java:1913) >> > at java.lang.String.subSequence(String.java:1946) >> > at java.util.regex.Matcher.getSubSequence(Matcher.java:1245) >> > at java.util.regex.Matcher.group(Matcher.java:490) >> > at java.util.Formatter$FormatSpecifier.<init>(Formatter.java:2675) >> > at java.util.Formatter.parse(Formatter.java:2528) >> > at java.util.Formatter.format(Formatter.java:2469) >> > at java.util.Formatter.format(Formatter.java:2423) >> > at java.lang.String.format(String.java:2790) >> > at >> scala.collection.immutable.StringLike$class.format(StringLike.scala:266) >> > at scala.collection.immutable.StringOps.format(StringOps.scala:31) >> > at org.apache.spark.util.Utils$.getCallSite(Utils.scala:944) >> > at org.apache.spark.rdd.RDD.<init>(RDD.scala:1227) >> > at org.apache.spark.rdd.RDD.<init>(RDD.scala:83) >> > at >> > >> org.apache.spark.rdd.PartitionwiseSampledRDD.<init>(PartitionwiseSampledRDD.scala:47) >> > at org.apache.spark.rdd.RDD$$anonfun$randomSplit$1.apply(RDD.scala:378) >> > at org.apache.spark.rdd.RDD$$anonfun$randomSplit$1.apply(RDD.scala:377) >> > at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) >> > at scala.collection.Iterator$class.foreach(Iterator.scala:727) >> > at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) >> > at >> scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) >> > at >> scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) >> > at >> scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) >> > at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) >> > at scala.collection.AbstractIterator.to(Iterator.scala:1157) >> > at >> > >> scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) >> > at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) >> > at >> scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) >> > at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) >> > at org.apache.spark.rdd.RDD.randomSplit(RDD.scala:379) >> > >> > >> > >> > ________________________________ >> > >> > The information contained in this e-mail is confidential and/or >> proprietary >> > to Capital One and/or its affiliates. The information transmitted >> herewith >> > is intended only for use by the individual or entity to which it is >> > addressed. If the reader of this message is not the intended >> recipient, you >> > are hereby notified that any review, retransmission, dissemination, >> > distribution, copying or other use of, or taking of any action in >> reliance >> > upon this information is strictly prohibited. If you have received this >> > communication in error, please contact the sender and delete the >> material >> > from your computer. >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: user-unsubscr...@spark.apache.org >> For additional commands, e-mail: user-h...@spark.apache.org >> >>
