TL;DR
Check your warehouse AUTO_SUSPEND setting. The default is 10 minutes. For most workloads, 1-2 minutes works fine. A Medium warehouse left running 8 extra minutes per query session costs ~$1,200/month in wasted credits.
The Surprise
A client called us in a panic. Their Snowflake bill jumped from $8,000 to $22,000 in a single month. No new data, no new users, no new dashboards. What happened?
They'd promoted a new BI dashboard to production. The dashboard ran 15 queries when users opened it, then users would read the results for 10-20 minutes before running more queries. Normal behavior.
The problem: their warehouse was configured with the default 10-minute AUTO_SUSPEND. Every dashboard session kept a Medium warehouse running for an extra 8-10 minutes of idle time. Multiply by 50 users, 5 times per day, 22 business days—and you've burned through a lot of credits doing nothing.
The Math
| AUTO_SUSPEND | Idle Time/Session | Monthly Waste* |
|---|---|---|
| 10 min (default) | ~8 min | $1,200 |
| 5 min | ~3 min | $450 |
| 1 min | ~30 sec | $75 |
*Medium warehouse, 50 users, 5 sessions/day, 22 days/month, $3/credit
The Fix
One line of SQL:
ALTER WAREHOUSE analytics_wh SET AUTO_SUSPEND = 60;
That's 60 seconds. For most BI workloads, this is plenty. The warehouse resumes in 1-2 seconds when the next query arrives—users won't notice.
Exception: If your warehouse initializes large caches or has complex startup overhead, test before setting AUTO_SUSPEND below 5 minutes. Some ETL workloads benefit from longer suspend times.
Quick Audit
Check all your warehouses right now:
SELECT
name,
size,
auto_suspend,
auto_resume
FROM snowflake.account_usage.warehouses
WHERE deleted_on IS NULL
ORDER BY auto_suspend DESC;
Any warehouse with AUTO_SUSPEND > 300 (5 minutes) for BI workloads is worth investigating.
