# Scheduling SQL Warehouse

> Warehouse schedules issue a Stop and wait for it to land, but auto-resume means any query (a 2am dashboard refresh, a BI tool's metadata poll) restarts the warehouse and bills from that moment. A stop schedule therefore only sticks after the overnight query sources are found and silenced.

Source: https://zop.dev/integrations/databricks/databricks-sqlwarehouse/schedule
Updated: 2026-08-19

---

## StopAndWait, confirmed down

The schedule stops the warehouse and waits for the stopped state rather than firing and
hoping. At that point the DBU meter is off and, for classic warehouses, the underlying
compute is released. So far, the simplest story imaginable.

## Auto-resume is the plot twist

SQL warehouses wake on demand: the first query to arrive restarts a stopped warehouse
automatically. That is a feature during the day and a schedule-killer at night. A dashboard
set to refresh hourly, a BI tool polling metadata, a data-quality job running SELECTs at 3am:
each one resurrects the warehouse, which then bills until its auto-stop idle window closes,
only to be woken again by the next refresh. The stop cron fires successfully every evening
and the invoice barely moves.

## The audit that makes the schedule real

Before trusting the calendar, read the warehouse's query history for the off-hours window and
identify every overnight caller. Reschedule the refreshes into business hours, point
monitoring probes elsewhere, or move genuinely necessary nightly jobs to their own small
warehouse. The schedule's value equals the silence you can actually arrange.

## Auto-stop already exists, the calendar closes the evening

Warehouses carry a native auto-stop, often 10-60 minutes of idle. Where it is configured
tightly, the calendar adds little; where teams stretched it to hours to dodge cold starts, the
19:00 stop reclaims the whole evening in one move. The two mechanisms are complementary:
auto-stop handles gaps during the day, the schedule guarantees the night.

## Classic and serverless wake differently

A serverless warehouse resumes in seconds, so morning users barely notice the schedule
existed. Classic warehouses take minutes to come up, which argues for a scheduled morning
start ahead of the first analysts rather than letting the first query of the day eat the
cold start.

Either way the stop costs nothing to attempt. A warehouse already stopped by its idle timer is
simply confirmed down, and the two mechanisms coexist without conflict.