Cost Management and Resource Optimization
Where credits go
Warehouses consume credits by size and runtime (per-second billing with a 60-second minimum per start and resumption). Cloud services consume credits only above 10% of daily warehouse credits. Serverless features - Snowpipe, automatic clustering, search optimization, replication, dynamic table refreshes, serverless tasks - bill separately. Storage bills for active data plus Time Travel and Fail-safe bytes; cloning is cheap until data diverges.
Controls
- Auto-suspend and auto-resume tuned to workload gaps; a very short suspend can cost more in restarts than it saves.
- Resource monitors at account or warehouse level with notify and suspend actions.
- Multi-cluster warehouses scale out for concurrency; set min/max clusters and scaling policy.
- Budgets and ACCOUNT_USAGE for chargeback by warehouse, role, or query tag.
- Query Acceleration Service offloads scan-heavy queries.
Traps
- Assuming cloud services are always free.
- Setting auto-suspend to 60 seconds on an interactive warehouse used every few minutes.
- Forgetting that a suspended warehouse still leaves storage and Fail-safe costs.
- Confusing scaling up (larger warehouse) with scaling out (more clusters).
- Overlooking replication and failover costs.
Rule of thumb: right-size the warehouse to the query, right-cluster it to concurrency, and set the resource monitor before the workload, not after. Worked example: a BI warehouse runs 8 hours a day at Medium; auto-suspend at 5 minutes, max clusters 3, and a resource monitor at 80% notify and 100% suspend prevent runaway costs.