Feature Request / Improvement
The Alluxio-backed file system cache seems easy to use from my testing.
One gap is the ability to choose some subset of tables/schemas/filepaths to be cached, with others skipping the cache entirely. It seems the Alluxio already has a few features allowing this or similar, but they aren't exposed in Trino.
Use case
I have an Iceberg catalog which contains both:
- smaller, often-used tables that are more important
- large, infrequently queried tables
With everything in a catalog sharing a cache, all the important files get pushed out of the cache whenever there's a large read on one of the second set of tables. It's possible to workaround this by e.g. exposing multiple catalogs, but that adds more logic for consumers.
Feature Request / Improvement
The Alluxio-backed file system cache seems easy to use from my testing.
One gap is the ability to choose some subset of tables/schemas/filepaths to be cached, with others skipping the cache entirely. It seems the Alluxio already has a few features allowing this or similar, but they aren't exposed in Trino.
Use case
I have an Iceberg catalog which contains both:
With everything in a catalog sharing a cache, all the important files get pushed out of the cache whenever there's a large read on one of the second set of tables. It's possible to workaround this by e.g. exposing multiple catalogs, but that adds more logic for consumers.