How to configure Apache Spark random worker ports

问题:

I am using Apache Spark to run machine learning algorithms and other big data tasks. Previously, I was using spark cluster standalone mode running spark master and worker on the same machine. Now, I added multiple worker machines and due to a tight firewall, I have to edit the random port of worker. Can anyone help how to change random spark ports and tell me exactly what configuration file needs to be edited? I read the spark documentation and it says spark-defaults.conf should be configured but I don't know how I can configure this file for particularly changing random ports of spark.

回答1:

check here https://spark.apache.org/docs/latest/configuration.html#networking

In the "Networking" section, you can see some of the ports are by default random. You can set them to your choice like below:

val conf = new SparkConf() 
    .setMaster(master) 
    .setAppName("namexxx") 
    .set("spark.driver.port", "51810") 
    .set("spark.fileserver.port", "51811") 
    .set("spark.broadcast.port", "51812") 
    .set("spark.replClassServer.port", "51813") 
    .set("spark.blockManager.port", "51814") 
    .set("spark.executor.port", "51815") 

回答2:

Update for Spark 2.x

---

Some libraries have been rewritten from scratch and many legacy *.port properties are now obsolete (cf. SPARK-10997 / SPARK-20605 / SPARK-12588 / SPARK-17678 / etc)

For Spark 2.1, for instance, the port ranges on which the driver will listen for executor traffic are

• between spark.driver.port and spark.driver.port+spark.port.maxRetries
• between spark.driver.blockManager.port and spark.driver.blockManager.port+spark.port.maxRetries

And the port range on which the executors will listen for driver traffic and/or other executors traffic is

• between spark.blockManager.port and spark.blockManager.port+spark.port.maxRetries

The "maxRetries" property allows for running several Spark jobs in parallel; if the base port is already used, then the new job will try the next one, etc, unless the whole range is already used.

Source:
   https://spark.apache.org/docs/2.1.1/configuration.html#networking
   https://spark.apache.org/docs/2.1.1/security.html under "Configuring ports"