I Failed the Turing Test

Spark troubleshooting

Posted on Author VintaPosted in Big Data, DevOps, Machine Learning

Apache Spark 2.x Troubleshooting Guide
https://www.slideshare.net/jcmia1/a-beginners-guide-on-troubleshooting-spark-applications
https://www.slideshare.net/jcmia1/apache-spark-20-tuning-guide

Check your cluster UI to ensure that workers are registered and have sufficient resources

PYSPARK_DRIVER_PYTHON="jupyter" \
PYSPARK_DRIVER_PYTHON_OPTS="notebook --ip 0.0.0.0" \
pyspark \
--packages "org.xerial:sqlite-jdbc:3.16.1,com.github.fommil.netlib:all:1.1.2" \
--driver-memory 4g \
--executor-memory 20g \
--master spark://TechnoCore.local:7077
TaskSchedulerImpl: Initial job has not accepted any resources; check your cluster UI to ensure that workers are registered and have sufficient resources

可能是你指定的 --executor-memory 超過了 worker 的 memory。

你可以在 Spark Master UI http://localhost:8080/ 看到各個 worker 總共有多少 memory 可以用。如果每台 worker 可以用的 memory 容量不同，Spark 就只會選擇那些 memory 大於 --executor-memory 的 workers。

ref:
https://spoddutur.github.io/spark-notes/distribution_of_executors_cores_and_memory_for_spark_application

SparkContext was shut down

ERROR Executor: Exception in task 1.0 in stage 6034.0 (TID 21592)
java.lang.StackOverflowError
...
ERROR LiveListenerBus: SparkListenerBus has already stopped! Dropping event SparkListenerJobEnd(55,1494185401195,JobFailed(org.apache.spark.SparkException: Job 55 cancelled because SparkContext was shut down))

可能是 executor 的記憶體不夠，導致 Out Of Memory (OOM) 了。

ref:
http://stackoverflow.com/questions/32822948/sparkcontext-was-shut-down-while-running-spark-on-a-large-dataset

Container exited with a non-zero exit code 56 (or some other numbers)

WARN org.apache.spark.scheduler.cluster.YarnSchedulerBackend$YarnSchedulerEndpoint: Container marked as failed: container_1504241464590_0001_01_000002 on host: albedo-w-1.c.albedo-157516.internal. Exit status: 56. Diagnostics: Exception from container-launch.
Container id: container_1504241464590_0001_01_000002
Exit code: 56
Stack trace: ExitCodeException exitCode=56:
    at org.apache.hadoop.util.Shell.runCommand(Shell.java:972)
    at org.apache.hadoop.util.Shell.run(Shell.java:869)
    at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:1170)
    at org.apache.hadoop.yarn.server.nodemanager.DefaultContainerExecutor.launchContainer(DefaultContainerExecutor.java:236)
    at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:305)
    at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:84)
    at java.util.concurrent.FutureTask.run(FutureTask.java:266)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
    at java.lang.Thread.run(Thread.java:748)

Container exited with a non-zero exit code 56

可能是 executor 的記憶體不夠，導致 Out Of Memory (OOM) 了。

ref:
http://stackoverflow.com/questions/39038460/understanding-spark-container-failure

Exception in thread "main" java.lang.StackOverflowError

Exception in thread "main" java.lang.StackOverflowError
    at java.io.ObjectOutputStream$BlockDataOutputStream.setBlockDataMode(ObjectOutputStream.java:1786)
    at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1495)
    at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1432)
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1178)
    at java.io.ObjectOutputStream.defaultWriteFields(ObjectOutputStream.java:1548)
    at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1509)
    at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1432)
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1178)
    at java.io.ObjectOutputStream.defaultWriteFields(ObjectOutputStream.java:1548)
    at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1509)
    at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1432)
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1178)
    at java.io.ObjectOutputStream.writeObject(ObjectOutputStream.java:348)
    at scala.collection.immutable.List$SerializationProxy.writeObject(List.scala:468)
    at sun.reflect.GeneratedMethodAccessor10.invoke(Unknown Source)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    ...

解決辦法：

import org.apache.spark.ml.recommendation.ALS
import org.apache.spark.sql.SparkSession

val spark: SparkSession = SparkSession.builder().getOrCreate()
val sc = spark.sparkContext
sc.setCheckpointDir("./spark-data/checkpoint")

// 因為 sc.setCheckpointDir() 就會啟用 checkpoint 了
// 所以可以不用特別指定 checkpointInterval
val als = new ALS()
  .setCheckpointInterval(2)

ref:
https://stackoverflow.com/questions/31484460/spark-gives-a-stackoverflowerror-when-training-using-als
https://stackoverflow.com/questions/35127720/what-is-the-difference-between-spark-checkpoint-and-persist-to-a-disk

Randomness of hash of string should be disabled via PYTHONHASHSEED

解決辦法：

$ cd $SPARK_HOME
$ cp conf/spark-env.sh.template conf/spark-env.sh
$ echo "export PYTHONHASHSEED=42" >> conf/spark-env.sh

ref:
https://issues.apache.org/jira/browse/SPARK-13330

It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transforamtion

Exception: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transformation. SparkContext can only be used on the driver, not in code that it run on workers. For more information, see SPARK-5063.

因為 spark.sparkContext 只能在 driver program 裡存取，不能被 worker 存取（例如那些丟給 RDD 執行的 lambda function 或是 UDF 就是在 worker 上執行的）。

ref:
https://spark.apache.org/docs/latest/rdd-programming-guide.html#passing-functions-to-spark
https://engineering.sharethrough.com/blog/2013/09/13/top-3-troubleshooting-tips-to-keep-you-sparking/

Spark automatically creates closures:

  • for functions that run on RDDs at workers,
  • and for any global variables that are used by those workers.

One closure is send per worker for every task. Closures are one way from the driver to the worker.

ref:
https://gerardnico.com/wiki/spark/closure

Unable to find encoder for type stored in a Dataset

Unable to find encoder for type stored in a Dataset.  Primitive types (Int, String, etc) and Product types (case classes) are supported by importing spark.implicits._  Support for serializing other types will be added in future releases. someDF.as[SomeCaseClass]

解決辦法：

import spark.implicits._

yourDF.as[YourCaseClass]

ref:
https://stackoverflow.com/questions/38664972/why-is-unable-to-find-encoder-for-type-stored-in-a-dataset-when-creating-a-dat

Task not serializable

Caused by: java.io.NotSerializableException: Settings
Serialization stack:
    - object not serializable (class: Settings, value: Settings@2dfe2f00)
    - field (class: Settings$$anonfun$1, name: $outer, type: class Settings)
    - object (class Settings$$anonfun$1, <function1>)
Caused by: org.apache.spark.SparkException:
    Task not serializable at org.apache.spark.util.ClosureCleaner$.ensureSerializable(ClosureCleaner.scala:298)

通常是你在 closure functions 裡使用了 driver program 裡的某個 object，因為 Spark 會自動 serialize 那個被引用的 object 一起丟給 worker node 執行，所以如果那個 object 或是 class 沒辦法被 serialize，就會出現這個錯誤。

ref:
https://www.safaribooksonline.com/library/view/spark-the-definitive/9781491912201/ch04.html#user-defined-functions
http://www.puroguramingu.com/2016/02/26/spark-dos-donts.html
https://stackoverflow.com/questions/36176011/spark-sql-udf-task-not-serialisable
https://stackoverflow.com/questions/22592811/task-not-serializable-java-io-notserializableexception-when-calling-function-ou
https://databricks.gitbooks.io/databricks-spark-knowledge-base/content/troubleshooting/javaionotserializableexception.html
https://mp.weixin.qq.com/s/BT6sXZlHcufAFLgTONCHsg

如果你只有在 Databricks Notebook 裡遇到這個錯誤，因為 Notebook 的運作機制跟一般的 Spark application 稍微有點不同，你可以試試 package cell。

ref:
https://docs.databricks.com/user-guide/notebooks/package-cells.html

java.lang.IllegalStateException: Cannot find any build directories.

java.lang.IllegalStateException: Cannot find any build directories.
    at org.apache.spark.launcher.CommandBuilderUtils.checkState(CommandBuilderUtils.java:248)
    at org.apache.spark.launcher.AbstractCommandBuilder.getScalaVersion(AbstractCommandBuilder.java:240)
    at org.apache.spark.launcher.AbstractCommandBuilder.buildClassPath(AbstractCommandBuilder.java:194)
    at org.apache.spark.launcher.AbstractCommandBuilder.buildJavaCommand(AbstractCommandBuilder.java:117)
    at org.apache.spark.launcher.WorkerCommandBuilder.buildCommand(WorkerCommandBuilder.scala:39)
    at org.apache.spark.launcher.WorkerCommandBuilder.buildCommand(WorkerCommandBuilder.scala:45)
    at org.apache.spark.deploy.worker.CommandUtils$.buildCommandSeq(CommandUtils.scala:63)
    at org.apache.spark.deploy.worker.CommandUtils$.buildProcessBuilder(CommandUtils.scala:51)
    at org.apache.spark.deploy.worker.ExecutorRunner.org$apache$spark$deploy$worker$ExecutorRunner$$fetchAndRunExecutor(ExecutorRunner.scala:145)
    at org.apache.spark.deploy.worker.ExecutorRunner$$anon$1.run(ExecutorRunner.scala:73)

可能的原因是沒有設置 SPARK_HOME 或是你的 launch script 沒有讀到該環境變數。