← Back to Tutorials

4. Functional Architecture

Query Processing

When a query is submitted, Cloud Services parses and optimizes it, then dispatches execution to the Virtual Warehouse. The warehouse scans compressed micro-partitions in parallel using MPP (massively parallel processing).

Concurrency

Multiple warehouses can access the same data simultaneously without lock contention. Each warehouse operates independently with its own compute resources.

Caching

Cache TypeScopeDuration
Result CacheQuery results24 hours (or until data changes)
Warehouse CacheData files in SSDWhile warehouse is running
Metadata CacheObject metadataAlways fresh

Time Travel

-- Query data as of 1 hour ago
SELECT * FROM my_table AT(TIMESTAMP => DATEADD(hour, -1, CURRENT_TIMESTAMP));

-- Query data before a specific statement
SELECT * FROM my_table BEFORE(STATEMENT => '019...');
✏️ Exercise: Using Snowflake's TIME_TRAVEL feature, query a table as it existed 30 minutes ago. Create a clone of that historical snapshot.