Yes, the whole idea of sending “agents” to do processing is poor performing and things like snowflake and Trino, where queries go to already deployed code, run rings around it.
Furthermore, pyspark is by far the most popular and used spark, and it’s also got the absolute world-worst atrocious mechanical sympathy. Why?
Developer velocity trumps compute velocity any day?
(I want the niceness of python and the performance of eg firebolt. Why must I pick?)
(There is a general thing to get spark “off heap” and use generic query compute on the spark sql space, but it is miles behind those who start off there)
Furthermore, pyspark is by far the most popular and used spark, and it’s also got the absolute world-worst atrocious mechanical sympathy. Why?
Developer velocity trumps compute velocity any day?
(I want the niceness of python and the performance of eg firebolt. Why must I pick?)
(There is a general thing to get spark “off heap” and use generic query compute on the spark sql space, but it is miles behind those who start off there)