tgravescs commented on a change in pull request #24615: [SPARK-27488][CORE]
Driver interface to support GPU resources
URL: https://github.com/apache/spark/pull/24615#discussion_r286189602
##########
File path: core/src/main/scala/org/apache/spark/SparkContext.scala
##########
@@ -360,6 +363,48 @@ class SparkContext(config: SparkConf) extends Logging {
Utils.setLogLevel(org.apache.log4j.Level.toLevel(upperCased))
}
+ /**
+ * Checks to see if any resources (GPU/FPGA/etc) are available to the driver
by looking
+ * at and processing the spark.driver.resource.resourceName.addresses and
+ * spark.driver.resource.resourceName.discoveryScript configs. The configs
have to be
+ * present when the driver starts, setting them after startup does not work.
+ *
+ * If any resource addresses configs were specified then assume all
resources will be specified
+ * in that way. Otherwise use the discovery scripts to find the resources.
Users should
+ * not really be setting the addresses config directly and should not be
mixing methods
+ * for different types of resources since the addresses config is meant for
Standalone mode
+ * and other cluster managers should use the discovery scripts.
+ */
+ private def setupDriverResources(): Unit = {
+ // Only call getAllWithPrefix once and filter on those since there could
be a lot of spark
+ // configs.
+ val allDriverResourceConfs =
_conf.getAllWithPrefix(SPARK_DRIVER_RESOURCE_PREFIX)
+ val resourcesWithAddrsInConfs =
+ SparkConf.getConfigsWithSuffix(allDriverResourceConfs,
SPARK_RESOURCE_ADDRESSES_SUFFIX)
+
+ _resources = if (resourcesWithAddrsInConfs.nonEmpty) {
Review comment:
also note this matches the way the executor does it as well. For the
executor, either the worker in standalone mode passes in a resources file with
what it has allocated per executor or it doesn't pass a file and we use the
discovery scripts. Yarn and k8s I don't expect to have issues with this as
they will always use discovery Script. If we want to be able to support both
ways we should update both places. I don't run standalone mode much so if you
have a usecase this makes sense we can certainly change it.
----------------------------------------------------------------
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:
us...@infra.apache.org
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
