ADBC in production: zero-copy, non-zero problems

Both dbt and PowerBI, among others, have announced support and usage of ADBC (Arrow Database Connectivity) to power their connectors. When ADBC launched, the pitch was compelling: zero-copy columnar data movement, vendor neutrality, and Arrow-native performance. But then production happened...

Turns out creating a new connectivity standard was the easy part! The hard part is observability for queries that mysteriously fail, type mappings that work for 90% of schemas but explode on that one user's TIMESTAMP WITH TIMEZONE or Decimal values, driver discovery, and so on.

This talk shares what we've learned from ADBC getting used in production environments. We'll explore the maturation of the ADBC ecosystem: Driver manifests for discovery, connection profiles, type support evolution, and more. Expect war stories, concrete adoption guidance, and honest admissions of where the gaps still exist (and are being worked on!).