My best practice:- Keep the number of executors equal to the number of spark clients that your cluster is configured for. Go with 2 cores per executor. You will fetch optimum performance with consistent distribution of batches for processing as confirmed from Spark UI. ... View more