通过 spark-submit 在 Kubernetes 中找不到 Uber jar

Uber jar not found in Kubernetes via spark-submit

我有一个非常简单的 Spark 作业,但我无法让它在 Kubernetes 中运行。我得到的错误是:

>     19/10/03 14:59:51 WARN DependencyUtils: Local jar /opt/spark/work-dir/target/scala-2.11/ScalaTest-assembly-1.0.jar does
> not exist, skipping.
>     19/10/03 14:59:51 WARN SparkSubmit$$anon: Failed to load ScalaTest.
>     java.lang.ClassNotFoundException: ScalaTest
>       at java.net.URLClassLoader.findClass(URLClassLoader.java:382)
>       at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
>       at java.lang.ClassLoader.loadClass(ClassLoader.java:357)
>       at java.lang.Class.forName0(Native Method)
>       at java.lang.Class.forName(Class.java:348)
>       at org.apache.spark.util.Utils$.classForName(Utils.scala:238)
>       at org.apache.spark.deploy.SparkSubmit.org$apache$spark$deploy$SparkSubmit$$runMain(SparkSubmit.scala:806)
>       at org.apache.spark.deploy.SparkSubmit.doRunMain(SparkSubmit.scala:161)
>       at org.apache.spark.deploy.SparkSubmit.submit(SparkSubmit.scala:184)
>       at org.apache.spark.deploy.SparkSubmit.doSubmit(SparkSubmit.scala:86)
>       at org.apache.spark.deploy.SparkSubmit$$anon.doSubmit(SparkSubmit.scala:920)
>       at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:929)
>       at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)

项目结构:

project/build.properties
project/plugins.sbt
src/main/scala/ScalaTest.scala
Dockerfile
build.sbt

build.properties

sbt.version=1.2.8

plugins.sbt

addSbtPlugin("com.eed3si9n" % "sbt-assembly" % "0.14.6")
addSbtPlugin("net.virtual-void" % "sbt-dependency-graph" % "0.10.0-RC1")

ScalaTest.scala

import org.apache.spark.sql.SparkSession
import org.apache.spark.SparkContext

object ScalaTest {
  def main(args: Array[String]) {
    val spark = SparkSession.builder.appName("ScalaTest").config("spark.master", "local[*]").getOrCreate()

    import spark.implicits._

    println("hello")

  }
}

Dockerfile 这只是一个基于从 Spark 二进制文件中的 kubernetes 文件夹构建的包装图像。在构建此图像之前,我确保我 运行 sbt assembly 生成了 Uber jar。

FROM spark:latest

WORKDIR /opt/spark/work-dir

COPY target/scala-2.11/ScalaTest-assembly-1.0.jar target/scala-2.11/ScalaTest-assembly-1.0.jar

build.sbt

name := "ScalaTest"

version := "1.0"

scalaVersion := "2.11.12"

val sparkVersion = "2.4.4"

libraryDependencies ++= Seq(
    "org.apache.spark" % "spark-core_2.11" % sparkVersion % "provided",
    "org.apache.spark" % "spark-sql_2.11" % sparkVersion % "provided"
)

最后是我的spark-submit。在执行此操作之前,我将映像推送到 ECR 的注册表,以便 EKS 可以提取该映像。我还指出了 uber jar 我的图像中的位置。

~/spark-2.4.4-bin-hadoop2.7/bin/spark-submit \
    --master k8s://{K8S_ENDPOINT}:443 \
    --deploy-mode cluster \
    --name test-job \
    --conf spark.kubernetes.container.image={ECR_IMAGE}:latest \
    --conf spark.kubernetes.submission.waitAppCompletion=false \
    --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \
    --conf spark.kubernetes.driver.pod.name=test-job \
    --class ScalaTest \
    local:///opt/spark/work-dir/target/scala-2.11/ScalaTest-assembly-1.0.jar

另请注意,当我 运行 下面的命令(spark-submit 在我的容器中,本地)时,它按预期工作:

docker run --rm -it my-custom-image ../bin/spark-submit target/scala-2.11/ScalaTest-assembly-1.0.jar

更新 检查组装好的 uber jar,我可以看到 ScalaTest 的 class 在那里。

jar tf target/scala-2.11/ScalaTest-assembly-1.0.jar

...
ScalaTest$.class
ScalaTest.class
...

这个问题的解决方案是不将 jar 保留在工作目录中,而是保留在 jars 文件夹中。我没有查看文档,但可能这是一个可以更改的环境变量。无论如何,Dockerfile 应该如下所示:

FROM spark:latest

COPY target/scala-2.11/ScalaTest-assembly-1.0.jar /ops/spark/jars/ScalaTest-assembly-1.0.jar

然后相应地更改 spark-submit

~/spark-2.4.4-bin-hadoop2.7/bin/spark-submit \
    --master k8s://{K8S_ENDPOINT}:443 \
    --deploy-mode cluster \
    --name test-job \
    --conf spark.kubernetes.container.image={ECR_IMAGE}:latest \
    --conf spark.kubernetes.submission.waitAppCompletion=false \
    --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \
    --conf spark.kubernetes.driver.pod.name=test-job \
    --class ScalaTest \
    local:///opt/spark/jars/ScalaTest-assembly-1.0.jar