[SPARK-32411] GPU Cluster Fail

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Not A Problem
• Affects Version/s: 3.0.0
• Component/s: PySpark, Web UI

Environment

Ihave a Apache Spark 3.0 cluster consisting of machines with multiple nvidia-gpus and I connect my jupyter notebook to the cluster using pyspark,

• Assignee: Unassigned
• Reporter: Vinh Tran
• Created: 23/Jul/20 12:34
• Updated: 11/Sep/20 18:55
• Resolved: 23/Jul/20 16:26

Description

I'm having a difficult time getting a GPU cluster started on Apache Spark 3.0. It was hard to find documentation on this, but I stumbled on a NVIDIA github page for Rapids which suggested the following additional edits to the spark-defaults.conf:

spark.task.resource.gpu.amount 0.25
spark.executor.resource.gpu.discoveryScript ./usr/local/spark/getGpusResources.sh

I have a Apache Spark 3.0 cluster consisting of machines with multiple nvidia-gpus and I connect my jupyter notebook to the cluster using pyspark, however it results in the following error: 

Py4JJavaError: An error occurred while calling None.org.apache.spark.api.java.JavaSparkContext.
: org.apache.spark.SparkException: You must specify an amount for gpu
	at org.apache.spark.resource.ResourceUtils$.$anonfun$parseResourceRequest$1(ResourceUtils.scala:142)
	at scala.collection.immutable.Map$Map1.getOrElse(Map.scala:119)
	at org.apache.spark.resource.ResourceUtils$.parseResourceRequest(ResourceUtils.scala:142)
	at org.apache.spark.resource.ResourceUtils$.$anonfun$parseAllResourceRequests$1(ResourceUtils.scala:159)
	at scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:238)
	at scala.collection.mutable.ArraySeq.foreach(ArraySeq.scala:75)
	at scala.collection.TraversableLike.map(TraversableLike.scala:238)
	at scala.collection.TraversableLike.map$(TraversableLike.scala:231)
	at scala.collection.AbstractTraversable.map(Traversable.scala:108)
	at org.apache.spark.resource.ResourceUtils$.parseAllResourceRequests(ResourceUtils.scala:159)
	at org.apache.spark.SparkContext$.checkResourcesPerTask$1(SparkContext.scala:2773)
	at org.apache.spark.SparkContext$.org$apache$spark$SparkContext$$createTaskScheduler(SparkContext.scala:2884)
	at org.apache.spark.SparkContext.<init>(SparkContext.scala:528)
	at org.apache.spark.api.java.JavaSparkContext.<init>(JavaSparkContext.scala:58)
	at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
	at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
	at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
	at java.lang.reflect.Constructor.newInstance(Constructor.java:423)
	at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:247)
	at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
	at py4j.Gateway.invoke(Gateway.java:238)
	at py4j.commands.ConstructorCommand.invokeConstructor(ConstructorCommand.java:80)
	at py4j.commands.ConstructorCommand.execute(ConstructorCommand.java:69)
	at py4j.GatewayConnection.run(GatewayConnection.java:238)
	at java.lang.Thread.run(Thread.java:748)

After this, I then tried adding another line to the conf per the instructions which results in no errors, however when I log in to the Web UI at localhost:8080, under Running Applications, the state remains at waiting.

spark.task.resource.gpu.amount                  2
spark.executor.resource.gpu.discoveryScript    ./usr/local/spark/getGpusResources.sh
spark.executor.resource.gpu.amount              1