Re: spark.akka.frameSize stalls job in 1.1.0
Is it because countByValue or toArray put too much stress on the driver, if there are many unique words To me it is a typical word count problem, then you can solve it as follows (correct me if I am wrong) val textFile = sc.textFile(“file) val counts = textFile.flatMap(line = line.split( )).map(word = (word, 1)).reduceByKey((a, b) = a + b) counts.saveAsTextFile(“file”)//any way you don’t want to collect results to master, and instead putting them in file. Thanks. Zhan Zhang On Aug 16, 2014, at 9:18 AM, Jerry Ye jerr...@gmail.com wrote: The job ended up running overnight with no progress. :-( On Sat, Aug 16, 2014 at 12:16 AM, Jerry Ye jerr...@gmail.com wrote: Hi Xiangrui, I actually tried branch-1.1 and master and it resulted in the job being stuck at the TaskSetManager: 14/08/16 06:55:48 INFO scheduler.TaskSchedulerImpl: Adding task set 1.0 with 2 tasks 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:0 as TID 2 on executor 8: ip-10-226-199-225.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:0 as 28055875 bytes in 162 ms 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:1 as TID 3 on executor 0: ip-10-249-53-62.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:1 as 28055875 bytes in 178 ms It's been 10 minutes with no progress on relatively small data. I'll let it run overnight and update in the morning. Is there some place that I should look to see what is happening? I tried to ssh into the executor and look at /root/spark/logs but there wasn't anything informative there. I'm sure using CountByValue works fine but my use of a HashMap is only an example. In my actual task, I'm loading a Trie data structure to perform efficient string matching between a dataset of locations and strings possibly containing mentions of locations. This seems like a common thing, to process input with a relatively memory intensive object like a Trie. I hope I'm not missing something obvious. Do you know of any example code like my use case? Thanks! - jerry -- CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: spark.akka.frameSize stalls job in 1.1.0
Hi Zhan, Thanks for looking into this. I'm actually using the hash map as an example of the simplest snippet of code that is failing for me. I know that this is just the word count. In my actual problem I'm using a Trie data structure to find substring matches. On Sun, Aug 17, 2014 at 11:35 PM, Zhan Zhang zzh...@hortonworks.com wrote: Is it because countByValue or toArray put too much stress on the driver, if there are many unique words To me it is a typical word count problem, then you can solve it as follows (correct me if I am wrong) val textFile = sc.textFile(“file) val counts = textFile.flatMap(line = line.split( )).map(word = (word, 1)).reduceByKey((a, b) = a + b) counts.saveAsTextFile(“file”)//any way you don’t want to collect results to master, and instead putting them in file. Thanks. Zhan Zhang On Aug 16, 2014, at 9:18 AM, Jerry Ye jerr...@gmail.com wrote: The job ended up running overnight with no progress. :-( On Sat, Aug 16, 2014 at 12:16 AM, Jerry Ye jerr...@gmail.com wrote: Hi Xiangrui, I actually tried branch-1.1 and master and it resulted in the job being stuck at the TaskSetManager: 14/08/16 06:55:48 INFO scheduler.TaskSchedulerImpl: Adding task set 1.0 with 2 tasks 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:0 as TID 2 on executor 8: ip-10-226-199-225.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:0 as 28055875 bytes in 162 ms 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:1 as TID 3 on executor 0: ip-10-249-53-62.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:1 as 28055875 bytes in 178 ms It's been 10 minutes with no progress on relatively small data. I'll let it run overnight and update in the morning. Is there some place that I should look to see what is happening? I tried to ssh into the executor and look at /root/spark/logs but there wasn't anything informative there. I'm sure using CountByValue works fine but my use of a HashMap is only an example. In my actual task, I'm loading a Trie data structure to perform efficient string matching between a dataset of locations and strings possibly containing mentions of locations. This seems like a common thing, to process input with a relatively memory intensive object like a Trie. I hope I'm not missing something obvious. Do you know of any example code like my use case? Thanks! - jerry -- CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You.
Re: spark.akka.frameSize stalls job in 1.1.0
Not sure exactly how you use it. My understanding is that in spark it would be better to keep the overhead of driver as less as possible. Is it possible to broadcast trie to executors, do computation there and then aggregate the counters (??) in reduct phase? Thanks. Zhan Zhang On Aug 18, 2014, at 8:54 AM, Jerry Ye jerr...@gmail.com wrote: Hi Zhan, Thanks for looking into this. I'm actually using the hash map as an example of the simplest snippet of code that is failing for me. I know that this is just the word count. In my actual problem I'm using a Trie data structure to find substring matches. On Sun, Aug 17, 2014 at 11:35 PM, Zhan Zhang zzh...@hortonworks.com wrote: Is it because countByValue or toArray put too much stress on the driver, if there are many unique words To me it is a typical word count problem, then you can solve it as follows (correct me if I am wrong) val textFile = sc.textFile(“file) val counts = textFile.flatMap(line = line.split( )).map(word = (word, 1)).reduceByKey((a, b) = a + b) counts.saveAsTextFile(“file”)//any way you don’t want to collect results to master, and instead putting them in file. Thanks. Zhan Zhang On Aug 16, 2014, at 9:18 AM, Jerry Ye jerr...@gmail.com wrote: The job ended up running overnight with no progress. :-( On Sat, Aug 16, 2014 at 12:16 AM, Jerry Ye jerr...@gmail.com wrote: Hi Xiangrui, I actually tried branch-1.1 and master and it resulted in the job being stuck at the TaskSetManager: 14/08/16 06:55:48 INFO scheduler.TaskSchedulerImpl: Adding task set 1.0 with 2 tasks 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:0 as TID 2 on executor 8: ip-10-226-199-225.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:0 as 28055875 bytes in 162 ms 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Starting task 1.0:1 as TID 3 on executor 0: ip-10-249-53-62.us-west-2.compute.internal (PROCESS_LOCAL) 14/08/16 06:55:48 INFO scheduler.TaskSetManager: Serialized task 1.0:1 as 28055875 bytes in 178 ms It's been 10 minutes with no progress on relatively small data. I'll let it run overnight and update in the morning. Is there some place that I should look to see what is happening? I tried to ssh into the executor and look at /root/spark/logs but there wasn't anything informative there. I'm sure using CountByValue works fine but my use of a HashMap is only an example. In my actual task, I'm loading a Trie data structure to perform efficient string matching between a dataset of locations and strings possibly containing mentions of locations. This seems like a common thing, to process input with a relatively memory intensive object like a Trie. I hope I'm not missing something obvious. Do you know of any example code like my use case? Thanks! - jerry -- CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You. -- CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You.
Shuffle overlapping
Hi all, I'm reading the implementation of the shuffle in Spark. My understanding is that it's not overlapping with upstream stage. Is it helpful to overlap the computation of upstream stage w/ the shuffle (I mean the network copy, like in Hadoop)? If it is, is there any plan to implement it in the any version? --Z -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/Shuffle-overlapping-tp7902.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: Shuffle overlapping
I think there's some discussion of this at https://issues.apache.org/jira/browse/SPARK-2387 and https://github.com/apache/spark/pull/1328. - Josh On Mon, Aug 18, 2014 at 9:46 AM, zycodefish opensourcecodef...@gmail.com wrote: Hi all, I'm reading the implementation of the shuffle in Spark. My understanding is that it's not overlapping with upstream stage. Is it helpful to overlap the computation of upstream stage w/ the shuffle (I mean the network copy, like in Hadoop)? If it is, is there any plan to implement it in the any version? --Z -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/Shuffle-overlapping-tp7902.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: mvn test error
The exception indicates that the forked process doesn’t executed as expected, thus the test case *should* fail. Instead of replacing the exception with a logWarning, capturing and printing stdout/stderr of the forked process can be helpful for diagnosis. Currently the only information we have at hand is the process exit code, it’s hard to determine the reason why the forked process fails. On Tue, Aug 19, 2014 at 1:27 PM, scwf wangf...@huawei.com wrote: hi, all I notice that jenkins may also throw this error when running tests( https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/18688/ consoleFull). This is because in Utils.executeAndGetOutput our progress exitCode is not 0, may be we should logWarning here rather than throw a exception? Utils.executeAndGetOutput { val exitCode = process.waitFor() stdoutThread.join() // Wait for it to finish reading output if (exitCode != 0) { throw new SparkException(Process + command + exited with code + exitCode) } } any idea? On 2014/8/15 11:01, scwf wrote: env: ubuntu 14.04 + spark master buranch mvn -Pyarn -Phive -Phadoop-2.4 -Dhadoop.version=2.4.0 -DskipTests clean package mvn -Pyarn -Phadoop-2.4 -Phive test test error: DriverSuite: Spark assembly has been built with Hive, including Datanucleus jars on classpath - driver should exit after finishing *** FAILED *** SparkException was thrown during property evaluation. (DriverSuite.scala:40) Message: Process List(./bin/spark-class, org.apache.spark.DriverWithoutCleanup, local) exited with code 1 Occurred at table row 0 (zero based, not counting headings), which had values ( master = local ) SparkSubmitSuite: Spark assembly has been built with Hive, including Datanucleus jars on classpath - launch simple application with spark-submit *** FAILED *** org.apache.spark.SparkException: Process List(./bin/spark-submit, --class, org.apache.spark.deploy.SimpleApplicationTest, --name, testApp, --master, local, file:/tmp/1408015655220-0/testJar-1408015655220.jar) exited with code 1 at org.apache.spark.util.Utils$.executeAndGetOutput(Utils.scala:810) at org.apache.spark.deploy.SparkSubmitSuite.runSparkSubmit( SparkSubmitSuite.scala:311) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$14. apply$mcV$sp(SparkSubmitSuite.scala:291) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$14. apply(SparkSubmitSuite.scala:284) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$14. apply(SparkSubmitSuite.scala:284) at org.scalatest.Transformer$$anonfun$apply$1.apply( Transformer.scala:22) at org.scalatest.Transformer$$anonfun$apply$1.apply( Transformer.scala:22) at org.scalatest.OutcomeOf$class.outcomeOf(OutcomeOf.scala:85) at org.scalatest.OutcomeOf$.outcomeOf(OutcomeOf.scala:104) at org.scalatest.Transformer.apply(Transformer.scala:22) ... Spark assembly has been built with Hive, including Datanucleus jars on classpath - spark submit includes jars passed in through --jar *** FAILED *** org.apache.spark.SparkException: Process List(./bin/spark-submit, --class, org.apache.spark.deploy.JarCreationTest, --name, testApp, --master, local-cluster[2,1,512], --jars, file:/tmp/1408015659416-0/ testJar-1408015659471.jar,fi le:/tmp/1408015659472-0/testJar-1408015659513.jar, file:/tmp/1408015659415-0/testJar-1408015659416.jar) exited with code 1 at org.apache.spark.util.Utils$.executeAndGetOutput(Utils.scala:810) at org.apache.spark.deploy.SparkSubmitSuite.runSparkSubmit( SparkSubmitSuite.scala:311) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$15. apply$mcV$sp(SparkSubmitSuite.scala:305) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$15. apply(SparkSubmitSuite.scala:294) at org.apache.spark.deploy.SparkSubmitSuite$$anonfun$15. apply(SparkSubmitSuite.scala:294) at org.scalatest.Transformer$$anonfun$apply$1.apply( Transformer.scala:22) at org.scalatest.Transformer$$anonfun$apply$1.apply( Transformer.scala:22) at org.scalatest.OutcomeOf$class.outcomeOf(OutcomeOf.scala:85) at org.scalatest.OutcomeOf$.outcomeOf(OutcomeOf.scala:104) at org.scalatest.Transformer.apply(Transformer.scala:22) ... but only test the specific suite as follows will be ok: mvn -Pyarn -Phadoop-2.4 -Phive -DwildcardSuites=org.apache.spark.DriverSuite test it seems when run with mvn -Pyarn -Phadoop-2.4 -Phive test,the process with Utils.executeAndGetOutput started can not exited successfully (exitcode is not zero) anyone has idea for this? -- Best Regards Fei Wang - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Spark on YARN webui
Hi, We are running the snapshots (new spark features) on YARN and I was wondering if the webui is available on YARN mode... The deployment document does not mention webui on YARN mode... Is it available ? Thanks. Deb