Details
-
Bug
-
Status: Resolved
-
Minor
-
Resolution: Fixed
-
2.2.0
-
None
Description
The executor tab on Spark UI page shows task as completed when an executor process that is running that task is killed using the kill command.
Steps:
1. Run a big Spark job. As an example, I ran a pyspark job with the following command:
$SPARK_HOME/bin/spark-submit --master yarn --deploy-mode cluster --queue default --num-executors 10 --driver-memory 2G --conf spark.pyspark.driver.python=./Python3/bin/python --conf spark.pyspark.python=./Python3/bin/python --archives hdfs:///user/USERNAME/Python3.zip#Python3 ~/pi.py
2. Go to the UI to see which executors are running.
3. Do an ssh to each of the executor hosts and kill the java process running on the respective port mentioned in the UI using the following command:
kill <pid> OR kill -9 <pid>