Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Duplicate
-
3.0.0, 3.0.1
-
None
-
None
Description
In MesosCluster mode, all of the spark jobs fail to run because by default "--py-files" is set to an empty string which causes spark-submit to use wrong jar name. This issue is caused by SPARK-26466
For ex:
--total-executor-cores
2
--py-files
--conf
spark.driver.maxResultSize=15g
so the very next --conf is used as the value for --py-files so `spark.driver.maxResultSize=15g` is used as jar name
which causes error as below:
20/10/19 20:19:18 WARN DependencyUtils: Local jar {dir}/slaves/d9971b08-4929-4b00-9677-b14088c38603-S13/frameworks/86ff1f10-79fd-44f0-b807-71741091b457-0000/executors/driver-20201019200043-0014-retry-9/runs/3697919f-c3dc-4a13-8928-2fb3a20ac98d/spark.driver.maxResultSize=15g does not exist, skipping.
20/10/19 20:19:18 WARN SparkSubmit$$anon$2: Failed to load org.apache.spark.examples.SparkPi.
java.lang.ClassNotFoundException: org.apache.spark.examples.SparkPi
In order to reproduce the bug run any spark provided examples on mesos in cluster mode:
./bin/spark-submit --class org.apache.spark.examples.SparkPi --master mesos://<server>:7077 --deploy-mode cluster --conf spark.master.rest.enabled=true --executor-memory 5G --total-executor-cores 2 <home>/examples/jars/spark-examples_2.12-3.0.1.jar 100