optimizing-snowflake-workloadslisted

# Optimizing Snowflake Workloads ## When to use - Snowflake queries are slow, queue, or spill to remote storage. - Credit/cost is higher than expected. - Choosing warehouse size, multi-cluster settings, or clustering keys. - Do NOT use for writing the SQL logic itself (use `optimizing-sql-queries`). ## Workflow ``` - [ ] Open the Query Profile: check partitions scanned, spilling, and pruning - [ ] Right-size: scale UP for heavy single queries, OUT for concurrency - [ ] Set auto-suspend low and auto-resume on - [ ] Add clustering only to very large, selectively-filtered tables - [ ] Verify with ACCOUNT_USAGE / WAREHOUSE_METERING history ``` 1. **Read the Query Profile.** Key signals: "Partitions scanned vs total" (pruning), "Bytes spilled to local/remote storage" (warehouse too small), and the most expensive operator. 2. **Right-size the warehouse.** Scale **up** (bigger size) for one heavy query that spills; scale **out** (multi-cluster) for many concurrent queries that queue. Bigger warehouses cost more per second but can be cheaper if they finish proportionally faster. 3. **Auto-suspend + auto-resume** — you pay per running second. Set auto-suspend to ~60s so idle warehouses stop. 4. **Clustering keys** only on large tables filtered by a high-cardinality column; check `SYSTEM$CLUSTERING_INFORMATION` and beware reclustering cost. ## Patterns **Right-sizing decision:** spilling to remote storage → scale up; queries `QUEUED` under concurrency → add