Snowflake Data Lakehouse Architecture

The Data Lakehouse architecture represents a paradigm shift in data platform design, unifying the best aspects of data lakes and data warehouses into a single, cohesive platform.

Architecture Diagram 2: Data Lakehouse Unified Architecture

What is a Data Lakehouse?

A Data Lakehouse combines the low-cost, flexible storage of data lakes with the performance, reliability, and ACID transactions of data warehouses.

Key Characteristics

What is a Data Lakehouse?

• Combines data lake flexibility with data warehouse reliability
• Open formats (Parquet, ORC) enable multi-engine access
• ACID transactions ensure data integrity
• Cost-effective storage with warehouse-grade performance

Snowflake's Lakehouse Implementation

External Tables

• Query data stored externally without copying
• Supports Parquet, JSON, ORC formats
• Auto-refresh for up-to-date queries

Iceberg Tables

• Open table format with ACID transactions
• Schema evolution without table recreation
• Time travel on Iceberg snapshots

Hybrid Tables

• OLTP + OLAP on the same table
• Primary key enforcement
• ACID transactions for single-row operations

Medallion Architecture

• Progressive data refinement — each layer adds quality and logic
• ACID transactions ensure integrity across layers

Cost Optimization

Key Takeaway: The Lakehouse pattern is ideal for organizations needing both lake flexibility and warehouse reliability. Snowflake's compute-storage separation makes this seamless.

Snowflake's Lakehouse Implementation

Snowflake implements the Lakehouse pattern through several key features:

External Tables

External tables allow Snowflake to query data stored externally without copying it:

-- Create external table from S3
CREATE OR REPLACE EXTERNAL TABLE my_external_table
  WITH LOCATION = @my_s3_stage
  FILE_FORMAT = (TYPE = PARQUET)
  AUTO_REFRESH = true;

-- Query external data directly
SELECT * FROM my_external_table
WHERE date >= '2024-01-01';

Iceberg Tables (Snowflake Native Iceberg)

Snowflake supports Apache Iceberg for open table formats:

-- Create Iceberg table
CREATE OR REPLACE EXTERNAL TABLE my_iceberg_table
  CATALOG = 'SNOWFLAKE'
  EXTERNAL_VOLUME = 'my_volume'
  TABLE_FORMAT = ICEBERG
  DATA_FILES_FORMAT = (TYPE = PARQUET);

-- Time travel on Iceberg tables
SELECT * FROM my_iceberg_table
  AT (OFFSET => -3600);  -- 1 hour ago

Hybrid Tables

Hybrid tables combine transactional and analytical workloads:

-- Create hybrid table for OLTP-like operations
CREATE OR REPLACE HYBRID TABLE orders (
  order_id INTEGER PRIMARY KEY,
  customer_id INTEGER,
  order_date TIMESTAMP_NTZ,
  total_amount DECIMAL(10,2),
  status VARCHAR(20)
);

-- ACID transactions on hybrid tables
BEGIN TRANSACTION;
  INSERT INTO orders VALUES (1, 100, CURRENT_TIMESTAMP(), 99.99, 'PENDING');
  UPDATE orders SET status = 'COMPLETED' WHERE order_id = 1;
COMMIT;

Data Lakehouse Patterns in Snowflake

Medallion Architecture

The Medallion (Multi-Hop) architecture organizes data into bronze, silver, and gold layers:

-- Bronze Layer: Raw data ingestion
CREATE OR REPLACE TABLE bronze.raw_orders (
  raw_data VARIANT
);

-- Silver Layer: Cleaned and validated data
CREATE OR REPLACE TABLE silver.orders AS
  SELECT
    raw_data:order_id::INTEGER AS order_id,
    raw_data:customer_id::INTEGER AS customer_id,
    raw_data:order_date::DATE AS order_date,
    raw_data:amount::DECIMAL(10,2) AS amount,
    CURRENT_TIMESTAMP() AS ingested_at
  FROM bronze.raw_orders;

-- Gold Layer: Business-ready aggregates
CREATE OR REPLACE TABLE gold.daily_sales_summary AS
  SELECT
    order_date,
    COUNT(*) AS total_orders,
    SUM(amount) AS total_revenue,
    AVG(amount) AS avg_order_value
  FROM silver.orders
  GROUP BY order_date;

Data Sharing Across Domains

-- Share data across business domains
CREATE SHARE sales_share;
GRANT USAGE ON DATABASE analytics_db TO SHARE sales_share;
GRANT USAGE ON SCHEMA gold TO SHARE sales_share;
GRANT SELECT ON gold.daily_sales_summary TO SHARE sales_share;

-- Consumers can query shared data
CREATE DATABASE sales_from_share FROM SHARE sales_share;
SELECT * FROM sales_from_share.gold.daily_sales_summary;

Performance Optimization for Lakehouse

Partitioning Strategies

-- Partition external data by date
CREATE OR REPLACE EXTERNAL TABLE daily_events
  PARTITION BY (event_date)
  WITH LOCATION = @events_stage
  FILE_FORMAT = (TYPE = PARQUET);

-- Query optimization through partition pruning
SELECT * FROM daily_events
  WHERE event_date = '2024-01-15';

Clustering for Large Tables

-- Cluster large fact tables
CREATE OR REPLACE TABLE large_fact_table CLUSTER BY (event_date, region)
  AS SELECT * FROM staging_events;

-- Manual re-clustering
ALTER TABLE large_fact_table RECLUSTER;

Cost Optimization in Lakehouse