Those logs you included are from the Spark executor processes, as opposed to the YARN NodeManager processes.
If you don't think you have access to the NodeManager logs, I would try setting spark.yarn.executor.memoryOverhead to something like 1024 or 2048 and seeing if that helps. If it does, it's because YARN was killing the containers. -Sandy On Thu, Oct 2, 2014 at 6:48 AM, Mike Bernico <mike.bern...@gmail.com> wrote: > Hello Xiangrui and Sandy, > > Thanks for jumping in to help. > > So, first thing... After my email last night I reran my code using 10 > executors, 2G each, and everything ran okay. So, that's good, but I'm > still curious as to what I was doing wrong. > > For Xiangrui's questions: > > My training set is 49174 observations x 61497 terms in a sparse vector > from spark's tf/idf transform. The partition size is 1025, which isn't > something I've tuned, I'm guessing it's related to input splits. I've > never called coalesce, etc. > > For Sandy's: > > I do not see any memory errors in the yarn logs other than this > occasionally: > > 14/10/01 19:25:54 INFO storage.MemoryStore: Will not store rdd_11_195 as > it would require dropping another block from the same RDD > 14/10/01 19:25:54 WARN spark.CacheManager: Not enough space to cache > partition rdd_11_195 in memory! Free memory is 236314377 bytes. > 14/10/01 19:25:57 INFO executor.Executor: Finished task 195.0 in stage 2.0 > (TID 1220). 1134 bytes result sent to driver > > The only other badness I see in those logs is: > > 14/10/01 19:40:35 INFO network.SendingConnection: Initiating connection to > [<hostname removed> :57359 > <http://rpl0000001273.opr.etlab.test.statefarm.org/10.233.51.34:57359>] > 14/10/01 19:40:35 WARN network.SendingConnection: Error finishing > connection to <hostname removed>:57359 > <http://rpl0000001273.opr.etlab.test.statefarm.org/10.233.51.34:57359> > java.net.ConnectException: Connection refused > at sun.nio.ch.SocketChannelImpl.checkConnect(Native Method) > at > sun.nio.ch.SocketChannelImpl.finishConnect(SocketChannelImpl.java:701) > at > org.apache.spark.network.SendingConnection.finishConnect(Connection.scala:313) > at > org.apache.spark.network.ConnectionManager$$anon$8.run(ConnectionManager.scala:226) > at > java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110) > at > java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603) > at java.lang.Thread.run(Thread.java:722) > > > I'm guessing those are from after the executors have died their mysterious > death. I'm happy ot send you the entire log if you'd like. > > Thanks! > > > On Thu, Oct 2, 2014 at 2:02 AM, Sandy Ryza <sandy.r...@cloudera.com> > wrote: > >> Hi Mike, >> >> Do you have access to your YARN NodeManager logs? When executors die >> randomly on YARN, it's often because they use more memory than allowed for >> their YARN container. You would see messages to the effect of "container >> killed because physical memory limits exceeded". >> >> -Sandy >> >> On Wed, Oct 1, 2014 at 8:46 PM, Xiangrui Meng <men...@gmail.com> wrote: >> >>> The cost depends on the feature dimension, number of instances, number >>> of classes, and number of partitions. Do you mind sharing those >>> numbers? -Xiangrui >>> >>> On Wed, Oct 1, 2014 at 6:31 PM, Mike Bernico <mike.bern...@gmail.com> >>> wrote: >>> > Hi Everyone, >>> > >>> > I'm working on training mllib's Naive Bayes to classify TF/IDF >>> vectoried >>> > docs using Spark 1.1.0. >>> > >>> > I've gotten this to work fine on a smaller set of data, but when I >>> increase >>> > the number of vectorized documents I get hung up on training. The >>> only >>> > messages I'm seeing are below. I'm pretty new to spark and I don't >>> really >>> > know where to go next to troubleshoot this. >>> > >>> > I'm running spark in yarn like this: >>> > spark-shell --master yarn-client --executor-memory 7G --driver memory >>> 7G >>> > --num-executors 3 >>> > >>> > I have three workers, each with 64G of ram and 8 cores. >>> > >>> > >>> > >>> > scala> val model = NaiveBayes.train(training, lambda = 1.0) >>> > 14/10/01 19:40:34 ERROR YarnClientClusterScheduler: Lost executor 2 on >>> > rpl0000001273.<removed>: remote Akka client disassociated >>> > 14/10/01 19:40:34 WARN TaskSetManager: Lost task 195.0 in stage 5.0 >>> (TID >>> > 2940, rpl0000001273.<removed>): ExecutorLostFailure (executor lost) >>> > 14/10/01 19:40:34 WARN TaskSetManager: Lost task 190.0 in stage 5.0 >>> (TID >>> > 2782, rpl0000001272.<removed>): FetchFailed(BlockManagerId(2, >>> > rpl0000001273.<removed>, 57359, 0), shuffleId=1, mapId=0, reduceId=190) >>> > 14/10/01 19:40:35 WARN TaskSetManager: Lost task 195.1 in stage 5.0 >>> (TID >>> > 2941, rpl0000001272.<removed>): FetchFailed(BlockManagerId(2, >>> > rpl0000001273.<removed>, 57359, 0), shuffleId=1, mapId=0, reduceId=195) >>> > 14/10/01 19:40:36 WARN TaskSetManager: Lost task 185.0 in stage 5.0 >>> (TID >>> > 2780, rpl0000001277.<removed>): FetchFailed(BlockManagerId(2, >>> > rpl0000001273.<removed>, 57359, 0), shuffleId=1, mapId=0, reduceId=185) >>> > 14/10/01 19:46:24 ERROR YarnClientClusterScheduler: Lost executor 1 on >>> > rpl0000001272.<removed>: remote Akka client disassociated >>> > 14/10/01 19:46:24 WARN TaskSetManager: Lost task 78.0 in stage 5.1 (TID >>> > 3377, rpl0000001272.<removed>): ExecutorLostFailure (executor lost) >>> > 14/10/01 19:46:25 WARN TaskSetManager: Lost task 79.0 in stage 5.1 (TID >>> > 3378, rpl0000001273.<removed>): FetchFailed(BlockManagerId(1, >>> > rpl0000001272.<removed>, 60926, 0), shuffleId=1, mapId=5, reduceId=220) >>> > 14/10/01 19:46:25 WARN TaskSetManager: Lost task 78.1 in stage 5.1 (TID >>> > 3379, rpl0000001273.<removed>): FetchFailed(BlockManagerId(1, >>> > rpl0000001272.<removed>, 60926, 0), shuffleId=1, mapId=5, reduceId=215) >>> > 14/10/01 19:46:29 WARN TaskSetManager: Lost task 73.0 in stage 5.1 (TID >>> > 3372, rpl0000001277.<removed>): FetchFailed(BlockManagerId(1, >>> > rpl0000001272.<removed>, 60926, 0), shuffleId=1, mapId=9, reduceId=210) >>> > 14/10/01 19:57:27 ERROR YarnClientClusterScheduler: Lost executor 3 on >>> > rpl0000001277.<removed>: remote Akka client disassociated >>> > 14/10/01 19:57:27 WARN TaskSetManager: Lost task 177.0 in stage 5.2 >>> (TID >>> > 4015, rpl0000001277.<removed>): ExecutorLostFailure (executor lost) >>> > 14/10/01 19:57:27 ERROR ConnectionManager: Corresponding >>> SendingConnection >>> > to ConnectionManagerId(rpl0000001277.<removed>,41425) not found >>> > 14/10/01 19:57:30 WARN TaskSetManager: Lost task 182.0 in stage 5.2 >>> (TID >>> > 4020, rpl0000001272.<removed>): FetchFailed(BlockManagerId(3, >>> > rpl0000001277.<removed>, 41425, 0), shuffleId=1, mapId=2, reduceId=340) >>> > 14/10/01 19:57:30 WARN TaskSetManager: Lost task 177.1 in stage 5.2 >>> (TID >>> > 4022, rpl0000001272.<removed>): FetchFailed(BlockManagerId(3, >>> > rpl0000001277.<removed>, 41425, 0), shuffleId=1, mapId=2, reduceId=335) >>> > 14/10/01 19:57:36 WARN TaskSetManager: Lost task 183.0 in stage 5.2 >>> (TID >>> > 4021, rpl0000001273.<removed>): FetchFailed(BlockManagerId(3, >>> > rpl0000001277.<removed>, 41425, 0), shuffleId=1, mapId=8, reduceId=345) >>> > 14/10/01 20:20:22 ERROR YarnClientClusterScheduler: Lost executor 4 on >>> > rpl0000001273.<removed>: remote Akka client disassociated >>> > 14/10/01 20:20:22 WARN TaskSetManager: Lost task 527.0 in stage 5.3 >>> (TID >>> > 5159, rpl0000001273.<removed>): ExecutorLostFailure (executor lost) >>> > 14/10/01 20:20:23 WARN TaskSetManager: Lost task 517.0 in stage 5.3 >>> (TID >>> > 5149, rpl0000001272.<removed>): FetchFailed(BlockManagerId(4, >>> > rpl0000001273.<removed>, 51049, 0), shuffleId=1, mapId=6, reduceId=690) >>> > 14/10/01 20:20:23 WARN TaskSetManager: Lost task 527.1 in stage 5.3 >>> (TID >>> > 5160, rpl0000001272.<removed>): FetchFailed(BlockManagerId(4, >>> > rpl0000001273.<removed>, 51049, 0), shuffleId=1, mapId=5, reduceId=700) >>> > 14/10/01 20:20:25 WARN TaskSetManager: Lost task 522.0 in stage 5.3 >>> (TID >>> > 5154, rpl0000001277.<removed>): FetchFailed(BlockManagerId(4, >>> > rpl0000001273.<removed>, 51049, 0), shuffleId=1, mapId=5, reduceId=695) >>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: user-unsubscr...@spark.apache.org >>> For additional commands, e-mail: user-h...@spark.apache.org >>> >>> >> >