If you create SQL warehouses through Declarative Automation Bundles and leave out auto_stop_mins, the default isn’t what you get in the UI. Bundles default to 10 minutes for Serverless and 120 for Pro/Classic. The UI defaults to 10 for Serverless and 45 for Pro/Classic.
Apparently someone deployed a Serverless warehouse with a bundle, didn’t set the field, and got 120 minutes instead of the 10 the docs list. That hurts if you have a bunch of short scheduled alerts, because each one wakes the warehouse up and then it just sits there until auto-stop kicks in.
A few other things from the same writeup:
The Serverless UI won’t go below 5 min auto-stop, but the API, CLI and bundles accept 1 (databricks warehouses edit --auto-stop-mins 1). Pro and Classic bottom out at 10.
A single slow or spilling query points to warehouse size. Queries waiting at capacity point to max clusters. Peak queued queries above zero doesn’t tell you which one on its own.
Power BI DirectQuery can send a query on every dashboard interaction. If the business only needs fresh data every few minutes, Import or Hybrid mode cuts a lot of that.
DBU rate alone is misleading. In one AWS US East example, Serverless is about $0.70/DBU, Pro $0.55 and Classic $0.22, but Pro and Classic bill infrastructure on top.
Is anyone running a 1-minute auto-stop on serverless for scheduled jobs? Do the cold starts bother you?
Full article : https://select.dev/blog/serverless-sql-warehouse-cost-optimization



