Snowflake Caching and Performance

Snowflake employs a multi-layered caching architecture that dramatically accelerates query performance by storing intermediate and final results at various levels.

Architecture Diagram 2: Multi-Layer Cache Lookup Flow

Result Cache

The result cache stores the final results of executed queries for reuse:

-- Enable result cache (default: ON)
ALTER SESSION SET USE_CACHED_RESULT = TRUE;

-- Check if query uses cache
SELECT /*+ NO_USE_CACHED_RESULT */
  order_date,
  COUNT(*) as order_count
FROM orders
GROUP BY order_date;

-- Monitor cache usage
SELECT
  query_id,
  query_text,
  result_cache_hit,
  execution_time_ms
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('hour', -1, CURRENT_TIMESTAMP())
))
WHERE query_text LIKE '%GROUP BY%';

Cache Invalidation Rules

Micro-Partition Cache

Micro-partitions are automatically cached after first access:

-- Check micro-partition cache statistics
SELECT
  table_name,
  partition_count,
  partitions_pruned,
  partitions_scanned
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('hour', -1, CURRENT_TIMESTAMP())
))
WHERE query_text LIKE '%orders%'
LIMIT 5;

-- Monitor cache hit ratio
SELECT
  table_name,
  total_micro_partitions,
  cached_micro_partitions,
  cache_hit_ratio
FROM TABLE(INFORMATION_SCHEMA.MICROPARTITION_CACHE_METRICS)
WHERE table_name = 'ORDERS';

Micro-Partition Pruning

-- Enable partition pruning (default: ON)
ALTER SESSION SET ENABLE_PRUNING_OPTIMIZATION = TRUE;

-- Check pruning effectiveness
EXPLAIN SELECT * FROM orders WHERE order_date = '2024-01-15';

-- Monitor pruning statistics
SELECT
  query_id,
  partitions_total,
  partitions_scanned,
  pruning_ratio
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('hour', -1, CURRENT_TIMESTAMP())
))
WHERE query_text LIKE '%WHERE order_date%';

Remote Disk Cache

Snowflake caches data on remote disk for persistent performance:

-- Check remote cache status
SELECT
  table_name,
  remote_disk_cache_bytes / 1024 / 1024 AS cache_size_mb,
  cache_hit_ratio
FROM INFORMATION_SCHEMA.REMOTE_DISK_CACHE_METRICS
WHERE table_name = 'ORDERS';

-- Force cache refresh
ALTER TABLE orders RECLUSTER;

Performance Optimization Strategies

Warehouse Sizing

-- Check warehouse performance
SELECT
  warehouse_name,
  avg_query_time_ms,
  total_queries,
  cache_hit_ratio
FROM TABLE(INFORMATION_SCHEMA.WAREHOUSE_METERING_HISTORY(
  START_TIME => DATEADD('day', -7, CURRENT_TIMESTAMP())
))
WHERE warehouse_name = 'COMPUTE_WH';

-- Scale warehouse for better caching
ALTER WAREHOUSE compute_wh SET WAREHOUSE_SIZE = 'LARGE';

Query Optimization

-- Use materialized views for complex aggregations
CREATE MATERIALIZED VIEW mv_daily_summary AS
SELECT
  order_date,
  COUNT(*) as order_count,
  SUM(amount) as total_revenue
FROM orders
GROUP BY order_date;

-- Enable result caching for specific queries
SELECT /*+ USE_CACHED_RESULT */
  order_date,
  COUNT(*) as order_count,
  SUM(amount) as total_revenue
FROM orders
GROUP BY order_date;

Partition Optimization

-- Cluster tables for better pruning
ALTER TABLE orders CLUSTER BY (order_date, region);

-- Check clustering depth
SELECT
  table_name,
  clustering_depth,
  clustering_information
FROM INFORMATION_SCHEMA.TABLE_STORAGE_METRICS
WHERE table_name = 'ORDERS';

-- Manual re-clustering
ALTER TABLE orders RECLUSTER;

Cache Monitoring

-- Comprehensive cache metrics
SELECT
  'Result Cache' as cache_type,
  SUM(result_cache_hit) as hits,
  COUNT(*) - SUM(result_cache_hit) as misses,
  ROUND(SUM(result_cache_hit) / COUNT(*) * 100, 2) as hit_ratio
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('day', -1, CURRENT_TIMESTAMP())
))

UNION ALL

SELECT
  'Micro-Partition Cache' as cache_type,
  SUM(partitions_cached) as hits,
  SUM(partitions_scanned) as misses,
  ROUND(SUM(partitions_cached) / SUM(partitions_scanned) * 100, 2) as hit_ratio
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('day', -1, CURRENT_TIMESTAMP())
));

Performance Best Practices

-- Monitor overall performance
SELECT
  query_id,
  query_text,
  execution_time_ms,
  bytes_scanned,
  result_cache_hit,
  ROUND(bytes_scanned / 1024 / 1024, 2) AS scanned_mb
FROM TABLE(INFORMATION_SCHEMA.QUERY_HISTORY(
  START_TIME => DATEADD('hour', -1, CURRENT_TIMESTAMP())
))
ORDER BY execution_time_ms DESC
LIMIT 10;

Summary

Key Takeaways

Result cache provides instant results for identical queries — zero execution time.

Micro-partition cache improves performance through intelligent data pruning.

Remote disk cache provides persistent performance benefits across sessions.

Cache hit ratio is a key performance metric — aim for high ratios.

Performance Optimization Checklist

1. Keep query text consistent to maximize result cache hits
2. Ensure sufficient warehouse size for parallel processing
3. Use clustering to improve partition pruning effectiveness
4. Monitor cache hit ratios across all cache layers
5. Right-size warehouses based on workload patterns
6. Use materialized views for complex, repeated aggregations