A query with multiple JOINs and aggregations runs slower than expected on a Photon cluster. After checking the Spark UI, some stages show 'photon=false'. What is the first diagnostic step?

Remaining questions and full explanations are VIP-exclusive.