Dockerfiles/spark/2.2.1

Deploy Spark Cluster of standalone mode

Master

docker service create \
	--name spark-master \
	--hostname spark-master \
	--network swarm-net \
	--replicas 1 \
	--detach true \
	--endpoint-mode dnsrr \
	newnius/spark:2.2.1 master

Slaves

docker service create \
	--name spark-slave \
	--network swarm-net \
	--replicas 5 \
	--detach true \
	--endpoint-mode dnsrr \
	newnius/spark:2.2.1 slave spark://spark-master:7077

Validate installation

spark-submit PI

spark-submit \
	--master spark://spark-master:7077 \
	--deploy-mode cluster \
	--class org.apache.spark.examples.JavaSparkPi \
	./examples/jars/spark-examples*.jar 100

spark-shell HDFS wordcount

Enter spark-shell --master spark://spark-master:7077 to enter shell.

val lines = sc.textFile("hdfs://hadoop-master:8020/user/root/input")

val words = lines.flatMap(_.split("\\s+"))

val wc = words.map(word => (word, 1)).reduceByKey(_ + _)

wc.collect()

val cnt = words.map(word => 1).reduce(_ + _)

Browse the web UI

You can expose the ports in the script, but I'd rather not since the slaves shoule occupy the same ports.

To access the web UI, deploy another (socks5) proxy to route the traffic.

If you don't one, try newnius/docker-proxy, it is rather easy to use.

Visit spark-master:8080 to view the cluster.