与Docker合作的Spark集群：BlockManagerId；本地类不兼容

Question

在使用Spark和Docker分发操作时，我遇到了类型不匹配的问题。我遵循的教程似乎很清楚。下面是我对Scala代码的尝试：

package test

import com.datastax.spark.connector.cql.CassandraConnector
import org.apache.spark.{SparkConf, SparkContext}
import readhub.sharedkernel.config.Settings

object Application extends App {
    import com.datastax.spark.connector._


    val conf = new SparkConf(true)
      .setAppName("Coordinator")
      .setMaster("spark://localhost:7077")
      .set("spark.cassandra.connection.host", "valid host")

    val sc = new SparkContext(conf)

    CassandraConnector(conf).withSessionDo { session =>
      session.execute("CREATE KEYSPACE test2 WITH REPLICATION = {'class': 'SimpleStrategy', 'replication_factor': 1 }")
      session.execute("CREATE TABLE test2.words (word text PRIMARY KEY, count int)")
      session.execute("INSERT INTO test2.words(word, count) VALUES('hey', 32)")

      sc.cassandraTable("test2", "words")
        .map(r => r.getString("word"))
        .foreach(process)

    }

  def process(word: String): Unit = {
    // Dummy processing
    println(word)
  }
} 

build.sbt是这样的：

import sbt.project

val sparkSql = "org.apache.spark" %% "spark-sql" % "2.3.0" % "provided"
val sparkCassandraConnector = "com.datastax.spark" %% "spark-cassandra-connector" % "2.3.0" % "provided"

lazy val commonSettings = Seq(
  version := "0.1",
  scalaVersion := "2.11.12",
  organization := "ch.heig-vd"
)

lazy val root = (project in file("."))
  .settings(
    commonSettings,
    name := "Root"
  )
  .aggregate(
    coordinator
  )

lazy val coordinator = project
  .settings(
    commonSettings,
    name := "Coordinator",
    libraryDependencies ++= Seq(
      sparkSql,
      sparkCassandraConnector
    )
  )

Dockerfile取自此图，并稍加修改，以使用Spark的2.3.0版：

FROM phusion/baseimage:0.9.22

ENV SPARK_VERSION 2.3.0
ENV SPARK_INSTALL /usr/local
ENV SPARK_HOME $SPARK_INSTALL/spark
ENV SPARK_ROLE master
ENV HADOOP_VERSION 2.7
ENV SPARK_MASTER_PORT 7077
ENV PYSPARK_PYTHON python3
ENV DOCKERIZE_VERSION v0.2.0

RUN apt-get update && \
    apt-get install -y openjdk-8-jdk autossh python3-pip && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

##### INSTALL DOCKERIZE
RUN curl -L -O https://github.com/jwilder/dockerize/releases/download/$DOCKERIZE_VERSION/dockerize-linux-amd64-$DOCKERIZE_VERSION.tar.gz && \
    tar -C /usr/local/bin -xzvf dockerize-linux-amd64-$DOCKERIZE_VERSION.tar.gz && \
    rm -rf dockerize-linux-amd64-$DOCKERIZE_VERSION.tar.gz

##### INSTALL APACHE SPARK WITH HDFS
RUN curl -s http://mirror.synyx.de/apache/spark/spark-$SPARK_VERSION/spark-$SPARK_VERSION-bin-hadoop$HADOOP_VERSION.tgz | tar -xz -C $SPARK_INSTALL && \
    cd $SPARK_INSTALL && ln -s spark-$SPARK_VERSION-bin-hadoop$HADOOP_VERSION spark

WORKDIR $SPARK_HOME

##### ADD Scripts
RUN mkdir /etc/service/spark
ADD runit/spark.sh /etc/service/spark/run
RUN chmod +x /etc/service/**/*

EXPOSE 4040 6066 7077 7078 8080 8081 8888

VOLUME ["$SPARK_HOME/logs"]

CMD ["/sbin/my_init"]

码头工人作曲。yml也很简单：

version: "3"

services:
  master:
    build: birgerk-apache-spark

    ports:
      - "7077:7077"
      - "8080:8080"

  slave:
    build: birgerk-apache-spark
    environment:
      - SPARK_ROLE=slave
      - SPARK_MASTER=master
    depends_on:
      - master

我将git repo克隆到文件夹Birgerk-apache-Spark中，只将Spark的版本更改为2.3.0。

最后，我用胶水粘合所有东西：

sbt coordinator/assembly

来制造脂肪罐

spark-submit --class test.Application --packages com.datastax.spark:spark-cassandra-connector_2.11:2.3.0 --master spark://localhost:7077 ReadHub\ Coordinator-assembly-0.1.jar

将jar提交到集群中。当我发出spark submit时出现错误：

错误TransportRequestHandler:199-在RPC id 7068633004064450609 java上调用RpcHandler#receive（）时出错。木卫一。InvalidClassException:org。阿帕奇。火花存储BlockManagerId；本地类不兼容：java上的stream classdesc serialVersionUID=6155820641931972169，本地类serialVersionUID=-3720498261147521051。io。对象流类。java上的initNonProxy（ObjectStreamClass.java:687）。伊奥。ObjectInputStream。java上的readNonProxyDesc（ObjectInputStream.java:1876）。伊奥。ObjectInputStream。java上的readClassDesc（ObjectInputStream.java:1745）。伊奥。ObjectInputStream。java上的readOrdinaryObject（ObjectInputStream.java:2033）。伊奥。ObjectInputStream。java上的readObject0（ObjectInputStream.java:1567）。伊奥。ObjectInputStream。java上的defaultReadFields（ObjectInputStream.java:2278）。伊奥。ObjectInputStream。readSerialData（ObjectInputStream.java:2202）[…]

从我的角度来看，Dockerfile可以正确地下载相应版本的Spark，该版本可以在我的build.sbt.中作为依赖项找到

我肯定我错过了一些基本的东西。谁能给我指出正确的方向吗？

非常感谢！

Answer 1

spark 2.3.3和spark 2.3.0之间的版本不匹配。

小心不要提交作业与SPARK_HOME定义在您的主机上这可能会导致这种问题