The job submit config is like --spark-executor-core 2, but when container runs on node, use top I can see the actually core consume is far more than 2, As shown below: