Long or Disabled Auto Stop on Databricks SQL Warehouses

For typical use Databricks recommends 45 minutes on Pro and Classic warehouses and 10 minutes on serverless warehouses, the UI defaults. Warehouses set well above those values, or with Auto Stop turned off, pay for long stretches with no running queries.

Longer settings usually come from two places. Teams raise the timer, or set it to 0, to avoid the multi-minute start of Pro and Classic warehouses, which Databricks' cost guidance notes leads users to accept the higher cost rather than stop warehouses during idle periods. And warehouses created through the SQL Warehouses API or infrastructure-as-code without an explicit value pick up the API default of 120 minutes rather than the UI default. BI tools that send periodic refresh or keep-alive queries also reset the timer, so a warehouse can stay up all day for a dashboard nobody is looking at.

Idle warehouse charges

A running warehouse bills DBUs, and for Pro and Classic also cloud instance charges, whether or not queries are running, until it stops

Auto Stop

Minutes with no running queries before the warehouse stops; 0 disables it

Recommended defaults

45 minutes for Pro and Classic (minimum 10) and 10 minutes for serverless (minimum 5 in the UI, as low as 1 via the API)

API default

Auto_stop_mins defaults to 120 minutes in the SQL Warehouses API when not set

• Query the latest record per warehouse in system.compute.warehouses and flag auto_stop_minutes = 0, Pro or Classic warehouses above 45, and serverless warehouses above 10
• Flag warehouses whose auto_stop_minutes is exactly 120, which usually means they were created by API or Terraform without setting the value
• In system.compute.warehouse_events, measure the time between the last query in system.query.history and the next STOPPING or STOPPED event for each warehouse to estimate idle minutes per day
• Group system.query.history by compute.warehouse_id, client_application and executed_by to find BI tools or service principals issuing periodic lightweight queries that keep warehouses from ever reaching Auto Stop
• Rank the resulting warehouses by DBUs in system.billing.usage (usage_metadata.warehouse_id) to prioritize

• Bring Auto Stop back to the recommended defaults or lower (45 minutes or down to 10 on Pro and Classic, 10 or down to 5 on serverless), weighing idle savings against more frequent cold starts
• Re-enable Auto Stop on any warehouse set to 0 unless it serves genuinely continuous traffic, and set auto_stop_mins explicitly in API calls and Terraform so new warehouses do not inherit the 120-minute default
• Move bursty interactive and BI workloads to serverless SQL warehouses, which start in seconds and make short Auto Stop values practical
• Reduce or schedule BI tool refreshes and remove keep-alive queries so dashboards do not hold warehouses open outside working hours
• For shared Pro or Classic warehouses that must be warm during business hours, keep a short timer and start the warehouse from a scheduled job or the SQL Warehouses API start call before the day begins, rather than a long timer that bills overnight

• Create a SQL warehousedocs.databricks.com
• Create a warehouse - SQL Warehouses APIdocs.databricks.com
• Best practices for cost optimizationdocs.databricks.com
• Warehouses system table referencedocs.databricks.com
• Warehouse events system table referencedocs.databricks.com
• Databricks Lakehouse Pricingdatabricks.com