zhouyejoe commented on a change in pull request #32007:
URL: https://github.com/apache/spark/pull/32007#discussion_r648874572
##########
File path: core/src/main/scala/org/apache/spark/util/Utils.scala
##########
@@ -2566,11 +2601,28 @@ private[spark] object Utils extends Logging {
}
/**
- * Push based shuffle can only be enabled when external shuffle service is
enabled.
+ * Push based shuffle can only be enabled when the application is submitted
+ * to run in YARN mode, with external shuffle service enabled and
+ * spark.yarn.maxAttempts or the yarn cluster default max attempts is set to
1.
+ * TODO: SPARK-35546 Support push based shuffle with multiple app attempts
*/
def isPushBasedShuffleEnabled(conf: SparkConf): Boolean = {
conf.get(PUSH_BASED_SHUFFLE_ENABLED) &&
- (conf.get(IS_TESTING).getOrElse(false) ||
conf.get(SHUFFLE_SERVICE_ENABLED))
+ (conf.get(IS_TESTING).getOrElse(false) ||
+ (conf.get(SHUFFLE_SERVICE_ENABLED) &&
+ conf.get(SparkLauncher.SPARK_MASTER, null) == "yarn") &&
+ getYarnMaxAttempts(conf) == 1)
+ }
+
+ /** Returns the maximum number of attempts to register the AM in YARN mode.
*/
+ def getYarnMaxAttempts(conf: SparkConf): Int = {
+ val sparkMaxAttempts =
conf.getOption("spark.yarn.maxAttempts").map(_.toInt)
+ val yarnMaxAttempts = getSparkOrYarnConfig(conf,
YarnConfiguration.RM_AM_MAX_ATTEMPTS,
+ YarnConfiguration.DEFAULT_RM_AM_MAX_ATTEMPTS.toString).toInt
+ sparkMaxAttempts match {
+ case Some(x) => if (x <= yarnMaxAttempts) x else yarnMaxAttempts
+ case None => yarnMaxAttempts
+ }
Review comment:
Added comment that this method will be removed after SPARK-35546
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
