I can see the query join operation takes ~80% of overall time and I during the run only 1 executor is really busy, can this be a situation that join is not parallelized and running on 1 executor only?
Can I find this out from profile information?
Have you tried with the data from Hash Aggregate Operator Metrics?
Hi @capochiani thanks for the reply. Yes I tried to use that kind of information
I could see there num_hash_partitions, but its still unclear if 1 partition per 1 executor or 1 partition per 1 cpu on the same executor?