I run spark-sql --master yarn command to lanuch a SparkSubmit process， when I registered a udf named AesCode and used it by select AesCode(name) from mytable, I found that the Executor(YarnCoarseGrainedExecutorBackend) jvm did not load the udf class while task is running on Executor machine. I did not understand that.
I think that the executor should load the class for the UDF before it can be used