How to setup Apache Spark and Zeppelin on Docker

Question

I'm trying to setup a Spark development environment with Zeppelin on Docker, but I'm having trouble connecting the Zeppelin and Spark containers.

I'm deploying a Docker Stack, with the current docker-compose

version: '3'
services:

  spark-master:
    image: gettyimages/spark
    command: bin/spark-class org.apache.spark.deploy.master.Master -h spark-master
    hostname: spark-master
    environment:
      SPARK_CONF_DIR: /conf
      SPARK_PUBLIC_DNS: 10.129.34.90
    volumes:
      - spark-master-volume:/conf
      - spark-master-volume:/tmp/data
    ports: 
      - 8000:8080

  spark-worker:
    image: gettyimages/spark
    command: bin/spark-class org.apache.spark.deploy.worker.Worker spark://spark-master:7077
    hostname: spark-worker
    environment:
      SPARK_MASTER_URL: spark-master:7077
      SPARK_CONF_DIR: /conf
      SPARK_PUBLIC_DNS: 10.129.34.90
      SPARK_WORKER_CORES: 2
      SPARK_WORKER_MEMORY: 2g
    volumes:
      - spark-worker-volume:/conf
      - spark-worker-volume:/tmp/data
    ports:
      - "8081-8100:8081-8100" 

  zeppelin:
    image: apache/zeppelin:0.8.0
    ports: 
      - 8080:8080
      - 8443:8443
    volumes:
      - spark-master-volume:/opt/zeppelin/logs
      - spark-master-volume:/opt/zeppelin/notebookcd
    environment:
      MASTER: "spark://spark-master:7077"
      SPARK_MASTER: "spark://spark-master:7077"
      SPARK_HOME: /usr/spark-2.4.1
    depends_on:
      - spark-master

volumes:
  spark-master-volume:
    driver: local
  spark-worker-volume:
    driver: local

It builds normally, but when I try to run Spark on Zeppelin, it throws me:

java.lang.RuntimeException: /zeppelin/bin/interpreter.sh: line 231: /usr/spark-2.4.1/bin/spark-submit: No such file or directory

I think the problem is in the volumes, but I can't get how to do it right.

Answer 1

Build a Dockerfile with content

FROM gettyimages/spark

ENV APACHE_SPARK_VERSION 2.4.1
ENV APACHE_HADOOP_VERSION 2.8.0
ENV ZEPPELIN_VERSION 0.8.1

RUN apt-get update 
RUN set -x \
    && curl -fSL "http://www-eu.apache.org/dist/zeppelin/zeppelin-0.8.1/zeppelin-0.8.1-bin-all.tgz" -o /tmp/zeppelin.tgz \
    && tar -xzvf /tmp/zeppelin.tgz -C /opt/ \
    && mv /opt/zeppelin-* /opt/zeppelin \
    && rm /tmp/zeppelin.tgz 

ENV SPARK_SUBMIT_OPTIONS "--jars /opt/zeppelin/sansa-examples-spark-2016-12.jar"
ENV SPARK_HOME "/usr/spark-2.4.1/"

WORKDIR /opt/zeppelin

CMD ["/opt/zeppelin/bin/zeppelin.sh"]

and then define your service within your docker-compose.yml file with prefix

version: '3'
services:
  zeppelin:
    build: ./zeppelin
    image: zeppelin:0.8.1-hadoop-2.8.0-spark-2.4.1
    ...

Finally, use docker-compose -f docker-compose.yml build to build the customised image before docker stack deploy

Answer 2

You need to install spark on your zeppelin docker instance to use spark-submit and update the spark interpreter config to point it to your spark cluster

zeppelin_notebook_server:
    container_name: zeppelin_notebook_server
    build:
      context: zeppelin/
    restart: unless-stopped
    volumes:
      - ./zeppelin/config/interpreter.json:/zeppelin/conf/interpreter.json:rw
      - ./zeppelin/notebooks:/zeppelin/notebook
      - ../sample-data:/sample-data:ro
    ports:
      - "8085:8080"
    networks:
      - general
    labels:
      container_group: "notebook"

  spark_base:
    container_name: spark-base
    build:
      context: spark/base
    image: spark-base:latest

  spark_master:
    container_name: spark-master
    build:
      context: spark/master/
    networks:
      - general
    hostname: spark-master
    ports:
      - "3030:8080"
      - "7077:7077"
    environment:
      - "SPARK_LOCAL_IP=spark-master"
    depends_on:
      - spark_base
    volumes:
      - ./spark/apps/jars:/opt/spark-apps
      - ./spark/apps/data:/opt/spark-data
      - ../sample-data:/sample-data:ro

  spark_worker_1:
    container_name: spark-worker-1
    build:
      context: spark/worker/
    networks:
      - general
    hostname: spark-worker-1
    ports:
      - "3031:8081"
    env_file: spark/spark-worker-env.sh
    environment:
      - "SPARK_LOCAL_IP=spark-worker-1"
    depends_on:
      - spark_master
    volumes:
      - ./spark/apps/jars:/opt/spark-apps
      - ./spark/apps/data:/opt/spark-data
      - ../sample-data:/sample-data:ro

  spark_worker_2:
    container_name: spark-worker-2
    build:
      context: spark/worker/
    networks:
      - general
    hostname: spark-worker-2
    ports:
      - "3032:8082"
    env_file: spark/spark-worker-env.sh
    environment:
      - "SPARK_LOCAL_IP=spark-worker-2"
    depends_on:
      - spark_master
    volumes:
      - ./spark/apps/jars:/opt/spark-apps
      - ./spark/apps/data:/opt/spark-data
      - ../sample-data:/sample-data:ro

Zeppelin docker file:

FROM "apache/zeppelin:0.8.1"

RUN wget http://apache.mirror.iphh.net/spark/spark-2.4.3/spark-2.4.3-bin-hadoop2.7.tgz --progress=bar:force && \
    tar xvf spark-2.4.3-bin-hadoop2.7.tgz && \
    mkdir -p /usr/local/spark && \
    mv spark-2.4.3-bin-hadoop2.7/* /usr/local/spark/. && \
    mkdir -p /sample-data

ENV SPARK_HOME "/usr/local/spark/"

Make sure your zeppelin spark interpreter config is same as: