Athena query costs spiking, partition projection isn't helping like expected

Rewrite those filters to compare the raw partition columns directly instead of wrapping them in CAST or DATE functions. That's almost always what's silently disabling partition pruning.

Check whether AQE is disabled in your Spark conf, skew join handling helped us a lot here.

Projection avoids the Glue Catalog partition-listing cost, but it does nothing about data scanned per query. If your queries aren't filtering on the partition columns in a form Athena can push down, you're still scanning everything.

Checked the query logs, a dashboard tool queries with date filters wrapped in a function, which defeats pruning.