175、Spark运维管理进阶之作业资源调度standalone

1. 启动external shuffle service
   每一个Worker运行的节点都得启动这个服务
   ./sbin/.start-shuffle-service.sh
2. 启动spark-shell，启用动态资源分配
   添加如下配置

--conf spark.shuffle.service.enabled=true \
--conf spark.dynamicAllocation.enabled=true \
--conf spark.shuffle.service.port=7337 \

spark-shell --master spark://spark-project-1:7077 --jars /opt/module/hive/lib/mysql-connector-java-5.1.34.jar --conf spark.shuffle.service.enabled=true --conf spark.dynamicAllocation.enabled=true --conf spark.shuffle.service.port=7337 --driver-memory 600m --executor-memory 600m --num-executors 1

3. 过60s，发现打印日志，说executor被removed，executor进程也没了
4. 然后动手写一个wordcount程序，最后提交job的时候，会动态申请一个新的executor，出来一个新的executor进程

sc.setLogLevel("INFO")
val lines = sc.textFile("hdfs://spark-project-1:9000/testdate/wordcount.txt")
val words = lines.flatMap(_.split(" "))
val paris = words.map((_,1))
val counts = paris.reduceByKey(_ + _)
counts.collect()

5. 然后整个作业执行完毕，证明external shuffle service+动态资源分配，流程可以走通
6. 再等60s，executor又被释放掉