spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Newberger <david.newber...@wandcorp.com>
Subject RE: About a problem running a spark job in a cdh-5.7.0 vmware image.
Date Fri, 03 Jun 2016 12:51:47 GMT
Alonso,

The CDH VM uses YARN and the default deploy mode is client. I’ve been able to use the CDH
VM for many learning scenarios.


http://www.cloudera.com/documentation/enterprise/latest.html
http://www.cloudera.com/documentation/enterprise/latest/topics/spark.html

David Newberger

From: Alonso [mailto:alonsoir@gmail.com]
Sent: Friday, June 3, 2016 5:39 AM
To: user@spark.apache.org
Subject: About a problem running a spark job in a cdh-5.7.0 vmware image.

Hi, i am developing a project that needs to use kafka, spark-streaming and spark-mllib, this
is the github project<https://github.com/alonsoir/awesome-recommendation-engine/tree/develop>.

I am using a vmware cdh-5.7-0 image, with 4 cores and 8 GB of ram, the file that i want to
use is only 16 MB, if i finding problems related with resources because the process outputs
this message:

                                   .set("spark.driver.allowMultipleContexts", "true")


<https://about.me/alonso.isidoro.roman?promo=email_sig&utm_source=email_sig&utm_medium=email_sig&utm_campaign=external_links>
16/06/03 11:58:09 WARN TaskSchedulerImpl: Initial job has not accepted any resources; check
your cluster UI to ensure that workers are registered and have sufficient resources



when i go to spark-master page, i can see this:


Spark Master at spark://192.168.30.137:7077

    URL: spark://192.168.30.137:7077
    REST URL: spark://192.168.30.137:6066 (cluster mode)
    Alive Workers: 0
    Cores in use: 0 Total, 0 Used
    Memory in use: 0.0 B Total, 0.0 B Used
    Applications: 2 Running, 0 Completed
    Drivers: 0 Running, 0 Completed
    Status: ALIVE

Workers
Worker Id Address State Cores Memory
Running Applications
Application ID Name Cores Memory per Node Submitted Time User State Duration
app-20160603115752-0001
(kill)
AmazonKafkaConnector 0 1024.0 MB 2016/06/03 11:57:52 cloudera WAITING 2.0 min
app-20160603115751-0000
(kill)
AmazonKafkaConnector 0 1024.0 MB 2016/06/03 11:57:51 cloudera WAITING 2.0 min


And this is the spark-worker output:

Spark Worker at 192.168.30.137:7078

    ID: worker-20160603115937-192.168.30.137-7078
    Master URL:
    Cores: 4 (0 Used)
    Memory: 6.7 GB (0.0 B Used)

Back to Master
Running Executors (0)
ExecutorID Cores State Memory Job Details Logs

It is weird isn't ? master url is not set up and there is not any ExecutorID, Cores, so on
so forth...

If i do a ps xa | grep spark, this is the output:

[cloudera@quickstart bin]$ ps xa | grep spark
 6330 ?        Sl     0:11 /usr/java/jdk1.7.0_67-cloudera/bin/java -cp /usr/lib/spark/conf/:/usr/lib/spark/lib/spark-assembly-1.6.0-cdh5.7.0-hadoop2.6.0-cdh5.7.0.jar:/etc/hadoop/conf/:/usr/lib/spark/lib/spark-assembly.jar:/usr/lib/hadoop/lib/*:/usr/lib/hadoop/*:/usr/lib/hadoop-hdfs/lib/*:/usr/lib/hadoop-hdfs/*:/usr/lib/hadoop-mapreduce/lib/*:/usr/lib/hadoop-mapreduce/*:/usr/lib/hadoop-yarn/lib/*:/usr/lib/hadoop-yarn/*:/usr/lib/hive/lib/*:/usr/lib/flume-ng/lib/*:/usr/lib/paquet/lib/*:/usr/lib/avro/lib/*
-Dspark.deploy.defaultCores=4 -Xms1g -Xmx1g -XX:MaxPermSize=256m org.apache.spark.deploy.master.Master

 6674 ?        Sl     0:12 /usr/java/jdk1.7.0_67-cloudera/bin/java -cp /etc/spark/conf/:/usr/lib/spark/lib/spark-assembly-1.6.0-cdh5.7.0-hadoop2.6.0-cdh5.7.0.jar:/etc/hadoop/conf/:/usr/lib/spark/lib/spark-assembly.jar:/usr/lib/hadoop/lib/*:/usr/lib/hadoop/*:/usr/lib/hadoop-hdfs/lib/*:/usr/lib/hadoop-hdfs/*:/usr/lib/hadoop-mapreduce/lib/*:/usr/lib/hadoop-mapreduce/*:/usr/lib/hadoop-yarn/lib/*:/usr/lib/hadoop-yarn/*:/usr/lib/hive/lib/*:/usr/lib/flume-ng/lib/*:/usr/lib/paquet/lib/*:/usr/lib/avro/lib/*
-Dspark.history.fs.logDirectory=hdfs:///user/spark/applicationHistory -Dspark.history.ui.port=18088
-Xms1g -Xmx1g -XX:MaxPermSize=256m org.apache.spark.deploy.history.HistoryServer

 8153 pts/1    Sl+    0:14 /usr/java/jdk1.7.0_67-cloudera/bin/java -cp /home/cloudera/awesome-recommendation-engine/target/pack/lib/*
-Dprog.home=/home/cloudera/awesome-recommendation-engine/target/pack -Dprog.version=1.0-SNAPSHOT
example.spark.AmazonKafkaConnector 192.168.1.35:9092 amazonRatingsTopic

 8413 ?        Sl     0:04 /usr/java/jdk1.7.0_67-cloudera/bin/java -cp /usr/lib/spark/conf/:/usr/lib/spark/lib/spark-assembly-1.6.0-cdh5.7.0-hadoop2.6.0-cdh5.7.0.jar:/etc/hadoop/conf/:/usr/lib/spark/lib/spark-assembly.jar:/usr/lib/hadoop/lib/*:/usr/lib/hadoop/*:/usr/lib/hadoop-hdfs/lib/*:/usr/lib/hadoop-hdfs/*:/usr/lib/hadoop-mapreduce/lib/*:/usr/lib/hadoop-mapreduce/*:/usr/lib/hadoop-yarn/lib/*:/usr/lib/hadoop-yarn/*:/usr/lib/hive/lib/*:/usr/lib/flume-ng/lib/*:/usr/lib/paquet/lib/*:/usr/lib/avro/lib/*
-Xms1g -Xmx1g -XX:MaxPermSize=256m org.apache.spark.deploy.worker.Worker spark://quickstart.cloudera:7077

 8619 pts/3    S+     0:00 grep spark

master is set up with four cores and 1 GB and worker has not any dedicated core and it is
using 1GB, that is weird isn't ? I have configured the vmware image with 4 cores (from eight)
and 8 GB (from 16).

This is how it looks my build.sbt:

libraryDependencies ++= Seq(
  "org.apache.kafka" % "kafka_2.10" % "0.8.1"
      exclude("javax.jms", "jms")
      exclude("com.sun.jdmk", "jmxtools")
      exclude("com.sun.jmx", "jmxri"),
   //not working play module!! check
   //jdbc,
   //anorm,
   //cache,
   // HTTP client
   "net.databinder.dispatch" %% "dispatch-core" % "0.11.1",
   // HTML parser
   "org.jodd" % "jodd-lagarto" % "3.5.2",
   "com.typesafe" % "config" % "1.2.1",
   "com.typesafe.play" % "play-json_2.10" % "2.4.0-M2",
   "org.scalatest" % "scalatest_2.10" % "2.2.1" % "test",
   "org.twitter4j" % "twitter4j-core" % "4.0.2",
   "org.twitter4j" % "twitter4j-stream" % "4.0.2",
   "org.codehaus.jackson" % "jackson-core-asl" % "1.6.1",
   "org.scala-tools.testing" % "specs_2.8.0" % "1.6.5" % "test",
   "org.apache.spark" % "spark-streaming-kafka_2.10" % "1.6.0-cdh5.7.0",
   "org.apache.spark" % "spark-core_2.10" % "1.6.0-cdh5.7.0",
   "org.apache.spark" % "spark-streaming_2.10" % "1.6.0-cdh5.7.0",
   "org.apache.spark" % "spark-sql_2.10" % "1.6.0-cdh5.7.0",
   "org.apache.spark" % "spark-mllib_2.10" % "1.6.0-cdh5.7.0",
   "com.google.code.gson" % "gson" % "2.6.2",
   "commons-cli" % "commons-cli" % "1.3.1",
   "com.stratio.datasource" % "spark-mongodb_2.10" % "0.11.1",
   // Akka
   "com.typesafe.akka" %% "akka-actor" % akkaVersion,
   "com.typesafe.akka" %% "akka-slf4j" % akkaVersion,
   // MongoDB
   "org.reactivemongo" %% "reactivemongo" % "0.10.0"
)

packAutoSettings

As you can see, i am using the exact version of spark modules for the pseudo cluster and i
want to use sbt-pack in order to create
an unix command, this is how i am declaring programmatically the spark context :


val sparkConf = new SparkConf().setAppName("AmazonKafkaConnector")
                                   //.setMaster("local[4]")
                                   .setMaster("spark://192.168.30.137:7077")
                                   .set("spark.cores.max", "2")

...

val ratingFile= "hdfs://192.168.30.137:8020/user/cloudera/ratings.csv"


println("Using this ratingFile: " + ratingFile)
  // first create an RDD out of the rating file
  val rawTrainingRatings = sc.textFile(ratingFile).map {
    line =>
      val Array(userId, productId, scoreStr) = line.split(",")
      AmazonRating(userId, productId, scoreStr.toDouble)
  }

  // only keep users that have rated between MinRecommendationsPerUser and MaxRecommendationsPerUser
products


//THIS IS THE LINE THAT PROVOKES the
WARN TaskSchedulerImp



!


val trainingRatings = rawTrainingRatings.groupBy(_.userId)
                                          .filter(r => MinRecommendationsPerUser <=
r._2.size  && r._2.size < MaxRecommendationsPerUser)
                                          .flatMap(_._2)
                                          .repartition(NumPartitions)
                                          .cache()

  println(s"Parsed $ratingFile. Kept ${trainingRatings.count()} ratings out of ${rawTrainingRatings.count()}")

My question is, do you see anything wrong with the code? is there anything terrible wrong
that i have to change? and,
what can i do to have this up and running with my resources?

What most annoys me is that the above code works perfectly in the console spark of the virtual
image but when I try to make it run
programmatically creating the unix with SBT-pack command does not work.

If the dedicated resources are too few to develop this project, what else can i do? i mean,
do i need to hire a tiny cluster with AWS
or any another provider? if that is a correct answer, which are yours recommendation?
Thank you very much for reading until here.

Regards,

Alonso




________________________________
View this message in context: About a problem running a spark job in a cdh-5.7.0 vmware image.<http://apache-spark-user-list.1001560.n3.nabble.com/About-a-problem-running-a-spark-job-in-a-cdh-5-7-0-vmware-image-tp27082.html>
Sent from the Apache Spark User List mailing list archive<http://apache-spark-user-list.1001560.n3.nabble.com/>
at Nabble.com.
Mime
View raw message