Single point of failure master node, to resolve this problem, it should help zookeeper, and start at least two master nodes to achieve highly reliable, relatively simple configuration:
Spark Cluster Planning: Master: hadoop01, hadoop04;
Worker：hadoop02、hadoop03、hadoop04
Zk cluster installation configuration, and start zk cluster (not repeat them here)
Stop spark all services, modify the configuration file spark-env.sh, delete the configuration file
And add the following configuration SPARK_MASTER_IP
export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=hadoop02,hadoop03,hadoop04 -Dspark.deploy.zookeeper.dir=/spark"
Distributed to hadoop02, hadoop03, under hadoop04 node
1. Modify the slaves on hadoop01 node configuration file specifies the contents of worker nodes
ps: if you modify the slaves nodes were also made to distribute the profile
2. First start zookeeper cluster
3. Do sbin / start-all.sh script on hadoop01, then sbin performed on hadoop04 / start-master.sh start a second Master
ps: If you use spark-shell start cluster configuration you need to add
spark-shell --master spark: // master01 : port1, master02: port2