Our team hit a familiar Snowflake paradox: slower ETL runs arrived at the same time as FinOps alerts about rising credits.

The warehouse in question was WH_ETL_BRONZE_01 , a multi-cluster warehouse dedicated to Bronze layer ingestion and merge workloads. What looked like a simple cost problem turned out to be a workload-isolation and concurrency problem.

The Problem

We started with this setup:

ALTER WAREHOUSE WH_ETL_BRONZE_01 SET WAREHOUSE_SIZE = 'XSMALL' , MAX_CLUSTER_COUNT = 10 , MIN_CLUSTER_COUNT = 1 , SCALING_POLICY = 'STANDARD' , AUTO_SUSPEND = 60 , MAX_CONCURRENCY_LEVEL = 8 ; -- default behavior

Enter fullscreen mode

Exit fullscreen mode

And we saw both of these at once:

Higher queue times ( queued_overload_time )

Higher daily credits

The critical issue was workload mix. Long-running MERGE statements (30 to 60 minutes) were running alongside tiny queries.…