snowflake-cost-tuning

# Snowflake Cost Tuning

## Overview

Optimize Snowflake costs through resource monitors, warehouse right-sizing, auto-suspend tuning, and credit consumption analysis.

## Snowflake Pricing Model

| Cost Component | What It Measures | Typical % of Bill |
|---------------|-----------------|-------------------|
| Compute (credits) | Warehouse running time | 60-80% |
| Storage | Data at rest (compressed) | 10-20% |
| Cloud services | Metadata ops, auth, compilation | 5-10% |
| Data transfer | Egress between regions/clouds | 0-5% |
| Serverless | Snowpipe, auto-clustering, MV refresh | Variable |

**Credit rates by warehouse size:**

| Size | Credits/Hour | Nodes |
|------|-------------|-------|
| X-Small | 1 | 1 |
| Small | 2 | 2 |
| Medium | 4 | 4 |
| Large | 8 | 8 |
| X-Large | 16 | 16 |
| 2X-Large | 32 | 32 |

## Instructions

### Step 1: Analyze Current Credit Consumption

```sql
-- Credits by warehouse (last 30 days)
SELECT warehouse_name,
       SUM(credits_used) AS total_credits,
       SUM(credits_used_compute) AS compute_credits,
       SUM(credits_used_cloud_services) AS cloud_credits,
       ROUND(SUM(credits_used) * 3.0, 2) AS est_cost_usd  -- ~$3/credit standard
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
WHERE start_time >= DATEADD(days, -30, CURRENT_TIMESTAMP())
GROUP BY warehouse_name
ORDER BY total_credits DESC;

-- Daily credit trend
SELECT DATE_TRUNC('day', start_time) AS day,
       SUM(credits_used) AS credits
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
WHERE start_time >= DATEADD(days, -30, CURRENT_TIMESTAMP())
GROUP BY day
ORDER BY day;

-- Idle warehouse time (credits wasted while no queries running)
SELECT warehouse_name,
       SUM(credits_used) AS total_credits,
       COUNT(DISTINCT query_id) AS queries,
       CASE WHEN COUNT(DISTINCT query_id) = 0 THEN SUM(credits_used)
            ELSE 0 END AS idle_credits
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY w
LEFT JOIN SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY q
  ON w.warehouse_name = q.warehouse_name
  AND DATE_TRUNC('hour', w.start_time) = DATE_TRUNC('hour', q.start_time)
WHERE w.start_time >= DATEADD(days, -7, CURRENT_TIMESTAMP())
GROUP BY warehouse_name
ORDER BY idle_credits DESC;
```

### Step 2: Set Up Resource Monitors

```sql
-- Account-level resource monitor
CREATE OR REPLACE RESOURCE MONITOR account_monthly
  WITH CREDIT_QUOTA = 5000
  FREQUENCY = MONTHLY
  START_TIMESTAMP = IMMEDIATELY
  TRIGGERS
    ON 50 PERCENT DO NOTIFY
    ON 75 PERCENT DO NOTIFY
    ON 90 PERCENT DO NOTIFY
    ON 100 PERCENT DO SUSPEND
    ON 110 PERCENT DO SUSPEND_IMMEDIATE;

ALTER ACCOUNT SET RESOURCE_MONITOR = account_monthly;

-- Per-warehouse monitor for ETL
CREATE OR REPLACE RESOURCE MONITOR etl_daily
  WITH CREDIT_QUOTA = 100
  FREQUENCY = DAILY
  START_TIMESTAMP = IMMEDIATELY
  TRIGGERS
    ON 80 PERCENT DO NOTIFY
    ON 100 PERCENT DO SUSPEND;

ALTER WAREHOUSE PROD_ETL_WH SET RESOURCE_MONITOR = etl_daily;
```

### Step 3: Optimize Auto-Suspend

```sql
-- Short auto-suspend for bursty workloads (ETL)
ALTER WAREHOUSE ETL_WH SET AUTO_SUSPEND = 60;     -- 1 minute

-- Longer for interactive analytics (avoids constant resume)
ALTER WAREHOUSE ANALYTICS_WH SET AUTO_SUSPEND = 300;  -- 5 minutes

-- Check current auto-suspend settings
SELECT name, size, auto_suspend, auto_resume,
       CASE WHEN auto_suspend > 300 THEN 'REVIEW: high auto_suspend'
            ELSE 'OK' END AS recommendation
FROM INFORMATION_SCHEMA.WAREHOUSES;

-- Never set auto_suspend = 0 in production (warehouse runs forever)
```

### Step 4: Right-Size Warehouses

```sql
-- Find oversized warehouses (low utilization)
SELECT warehouse_name, warehouse_size,
       AVG(avg_running) AS avg_queries_running,
       AVG(avg_queued_load) AS avg_queries_queued,
       CASE
         WHEN AVG(avg_queued_load) > 1 THEN 'SCALE UP or add clusters'
         WHEN AVG(avg_running) < 1 THEN 'Consider DOWNSIZE'
         ELSE 'RIGHT-SIZED'
       END AS recommendation
FROM TABLE(INFORMATION_SCHEMA.WAREHOUSE_LOAD_HISTORY(
  DATE_RANGE_START => DATEADD(days, -7, CURRENT_TIMESTAMP())
))
GROUP BY warehouse_name, warehouse_size;

-- Downsize underutilized warehouses
ALTER WAREHOUSE DEV_WH SET WAREHOUSE_SIZE = 'XSMALL';
```

### Step 5: Reduce Storage Costs

```sql
-- Find large unused tables
SELECT table_catalog, table_schema, table_name,
       bytes / 1e9 AS gb,
       row_count,
       last_altered,
       DATEDIFF('day', last_altered, CURRENT_DATE()) AS days_since_modified
FROM SNOWFLAKE.ACCOUNT_USAGE.TABLE_STORAGE_METRICS
WHERE bytes > 1e9  -- > 1 GB
  AND DATEDIFF('day', last_altered, CURRENT_DATE()) > 90
ORDER BY bytes DESC;

-- Reduce Time Travel retention for non-critical tables
ALTER TABLE staging.temp_data SET DATA_RETENTION_TIME_IN_DAYS = 1;

-- Use transient tables for staging (no Fail-safe storage)
CREATE TRANSIENT TABLE staging.temp_load (
  id INTEGER, data VARIANT
);
```

### Step 6: Serverless Feature Cost Control

```sql
-- Monitor Snowpipe costs
SELECT pipe_name, SUM(credits_used) AS credits
FROM SNOWFLAKE.ACCOUNT_USAGE.PIPE_USAGE_HISTORY
WHERE start_time >= DATEADD(days, -30, CURRENT_TIMESTAMP())
GROUP BY pipe_name
ORDER BY credits DESC;

-- Monitor auto-clustering costs
SELECT table_name, SUM(credits_used) AS credits
FROM SNOWFLAKE.ACCOUNT_USAGE.AUTOMATIC_CLUSTERING_HISTORY
WHERE start_time >= DATEADD(days, -30, CURRENT_TIMESTAMP())
GROUP BY table_name
ORDER BY credits DESC;

-- Monitor materialized view refresh costs
SELECT table_name, SUM(credits_used) AS credits
FROM SNOWFLAKE.ACCOUNT_USAGE.MATERIALIZED_VIEW_REFRESH_HISTORY
WHERE start_time >= DATEADD(days, -30, CURRENT_TIMESTAMP())
GROUP BY table_name
ORDER BY credits DESC;
```

## Cost Reduction Checklist

- [ ] Resource monitors on all production warehouses
- [ ] Auto-suspend < 5 minutes on all warehouses
- [ ] No `WAREHOUSE_SIZE > 'MEDIUM'` without justification
- [ ] Transient tables for staging/temp data
- [ ] Time Travel retention minimized for non-critical tables
- [ ] Clustering keys only on tables > 1TB with frequent filter queries
- [ ] Review serverless feature costs monthly

## Error Handling

| Issue | Cause | Solution |
|-------|-------|----------|
| Unexpected credit spike | Runaway query or always-on warehouse | Check QUERY_HISTORY, set auto-suspend |
| Resource monitor suspended warehouse | Exceeded quota | Increase quota or optimize workload |
| High cloud services cost | Many small metadata queries | Batch operations, reduce DDL frequency |
| Storage growing fast | No cleanup policy | Archive old data, use transient tables |

## Resources

- [Cost Controls](https://docs.snowflake.com/en/user-guide/cost-controlling-controls)
- [Resource Monitors](https://docs.snowflake.com/en/user-guide/resource-monitors)
- [CREATE RESOURCE MONITOR](https://docs.snowflake.com/en/sql-reference/sql/create-resource-monitor)
- [Budgets](https://docs.snowflake.com/en/user-guide/budgets)

## Next Steps

For architecture patterns, see `snowflake-reference-architecture`.