When we run a Spark cluster on EMR, we often create a new cluster for every Spark job. In that case, we want to use all available resources, but changing the configuration is annoying and error-prone. How many times have you forgotten to change the Spark settings after changing the EMR instances to less powerful ones? Of course, the Spark job failed because it could not allocate the resources you wanted.
What is even worse, when you forget to change the settings after changing the instance to a bigger one, you pay for a better cluster, but you are not using it entirely.
Fortunately, Spark’s EMR version has a special configuration parameter that replaces all of the cumbersome parameters, such as the executor memory, the executor cores, or parallelism.
Instead of them, we should enable the maximizeResourceAllocation feature:
--conf maximizeResourceAllocation=true
when we call the spark-submit script.
Want to build AI systems that actually work?
Download my expert-crafted GenAI Transformation Guide for Data Teams and discover how to properly measure AI performance, set up guardrails, and continuously improve your AI solutions like the pros.