How do I get the YARN ContainerId from inside the container?

问题

I'm running a Spark job on YARN and would like to get the YARN container ID (as part of a requirement to generate unique IDs across a set of Spark jobs). I can see the Container.getId() method to get the ContainerId but no idea how to get a reference to the current running container from YARN. Is this even possible? How does a YARN container get it's own information?

回答1:

The only way that I could get something was to use the logging directory. The following works in a spark shell.

import org.apache.hadoop.yarn.api.records.ContainerId

def f(): String = {
  val localLogDir: String = System.getProperty("spark.yarn.app.container.log.dir")
  val containerIdString: String = localLogDir.split("/").last
  val containerIdLong: Long = ContainerId.fromString(containerIdString).getContainerId
  containerIdLong.toHexString
}

val rdd1 = sc.parallelize((1 to 10)).map{ _ => f() }
rdd1.distinct.collect().foreach(println)

回答2:

here below description how spark store the Container ID

Spark hide the container id and expose the executor id per application/job so if you are planning to maintain the unique id per spark job, my suggestion to use application id which spark gives you then you can add your some string to make unique for you

below spark code from "YarnAllocator.scala"

private[yarn] val executorIdToContainer = new HashMap[String, Container]

回答3:

YARN will export all of the environment variables listed here: https://github.com/apache/hadoop/blob/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api/src/main/java/org/apache/hadoop/yarn/api/ApplicationConstants.java#L117

So you should be able to access it like:

sys.env.get(ApplicationConstants.Environment.CONTAINER_ID.toString)
// or, equivalently
sys.env.get("CONTAINER_ID")

来源：https://stackoverflow.com/questions/41724300/how-do-i-get-the-yarn-containerid-from-inside-the-container

标签

Hadoop

apache-spark

yarn