Snowflake SnowPro Core (COF-C03) Study Notes

Snowflake AI Data Cloud Architecture

Three-layer architecture, editions, virtual warehouses, and Snowpark

Domain Weight

Snowflake AI Data Cloud Features & Architecture is 25% of the COF-C03 exam — the highest-weighted domain.

Three-Layer Architecture

Database Storage Layer

Data stored in Snowflake-managed cloud storage as micro-partitions. Columnar, compressed, automatically organized. Independent of compute.

Query Processing Layer

Virtual warehouses (MPP compute clusters) execute queries. Multiple warehouses can run simultaneously against shared data without contention.

Cloud Services Layer

Infrastructure management, optimization, security, metadata. Always running — no compute credits charged unless >10% of compute.

Micro-Partitions

Automatic — Snowflake manages partition creation and pruning; users cannot manually partition
Columnar format — queries read only needed columns (column pruning)
Compressed — ~3-10x compression; actual size is compressed
Immutable — DML creates new partitions; old ones are replaced, enabling Time Travel
~50-500MB of uncompressed data each; stored as compressed blobs
Partition pruning: Cloud Services reads metadata to skip irrelevant partitions without scanning

Virtual Warehouses

Property	Details
Sizes	X-Small (1 node) → X-Large (16) → 2X (32) → 3X (64) → 4X (128) → 5X (256) → 6X-Large (512 nodes)
Multi-Cluster	Enterprise+: auto-scale from min to max clusters for concurrency. AUTO or CUSTOM scaling policy.
Auto-Suspend	Warehouse suspends after N seconds of inactivity; local disk cache is CLEARED on suspend
Auto-Resume	Warehouse resumes automatically on query submission (default enabled)
Credits	Charged per second, minimum 60 seconds. Starting a resumed warehouse does NOT charge min 60 again.
Concurrency Slots	Each warehouse has a fixed number of concurrency slots; queuing occurs when exceeded

Snowflake Editions

Edition	Key Additional Features
Standard	All core features, 1-day Time Travel, all file formats, Snowpipe
Enterprise	Multi-cluster warehouses, 90-day Time Travel, column-level security, materialized views, row access policies
Business Critical	HIPAA/PCI-DSS compliance, Tri-Secret Secure, Customer-Managed Keys, Private Link, enhanced security
Virtual Private Snowflake (VPS)	Dedicated Snowflake environment, highest isolation, government use cases

Snowpark & Cortex AI

Snowpark: DataFrame API for Python, Java, Scala running inside Snowflake compute
Code runs in Snowflake — no data movement to external systems
Supports: UDFs, stored procedures, ML model training and inference
Snowflake Cortex: LLM functions (COMPLETE, SUMMARIZE, SENTIMENT, TRANSLATE) built into SQL
ML Functions in Cortex: FORECAST, ANOMALY_DETECTION, CLASSIFICATION — no-code ML
Cortex runs serverlessly on Snowflake GPU infrastructure

Snowflake SnowPro Core (COF-C03) Study Notes

Snowflake AI Data Cloud Architecture

Three-layer architecture, editions, virtual warehouses, and Snowpark

Domain Weight

Snowflake AI Data Cloud Features & Architecture is 25% of the COF-C03 exam — the highest-weighted domain.

Three-Layer Architecture

Database Storage Layer

Data stored in Snowflake-managed cloud storage as micro-partitions. Columnar, compressed, automatically organized. Independent of compute.

Query Processing Layer

Virtual warehouses (MPP compute clusters) execute queries. Multiple warehouses can run simultaneously against shared data without contention.

Cloud Services Layer

Infrastructure management, optimization, security, metadata. Always running — no compute credits charged unless >10% of compute.

Micro-Partitions

Automatic — Snowflake manages partition creation and pruning; users cannot manually partition
Columnar format — queries read only needed columns (column pruning)
Compressed — ~3-10x compression; actual size is compressed
Immutable — DML creates new partitions; old ones are replaced, enabling Time Travel
~50-500MB of uncompressed data each; stored as compressed blobs
Partition pruning: Cloud Services reads metadata to skip irrelevant partitions without scanning

Virtual Warehouses

Property	Details
Sizes	X-Small (1 node) → X-Large (16) → 2X (32) → 3X (64) → 4X (128) → 5X (256) → 6X-Large (512 nodes)
Multi-Cluster	Enterprise+: auto-scale from min to max clusters for concurrency. AUTO or CUSTOM scaling policy.
Auto-Suspend	Warehouse suspends after N seconds of inactivity; local disk cache is CLEARED on suspend
Auto-Resume	Warehouse resumes automatically on query submission (default enabled)
Credits	Charged per second, minimum 60 seconds. Starting a resumed warehouse does NOT charge min 60 again.
Concurrency Slots	Each warehouse has a fixed number of concurrency slots; queuing occurs when exceeded

Snowflake Editions

Edition	Key Additional Features
Standard	All core features, 1-day Time Travel, all file formats, Snowpipe
Enterprise	Multi-cluster warehouses, 90-day Time Travel, column-level security, materialized views, row access policies
Business Critical	HIPAA/PCI-DSS compliance, Tri-Secret Secure, Customer-Managed Keys, Private Link, enhanced security
Virtual Private Snowflake (VPS)	Dedicated Snowflake environment, highest isolation, government use cases

Snowpark & Cortex AI

Snowpark: DataFrame API for Python, Java, Scala running inside Snowflake compute
Code runs in Snowflake — no data movement to external systems
Supports: UDFs, stored procedures, ML model training and inference
Snowflake Cortex: LLM functions (COMPLETE, SUMMARIZE, SENTIMENT, TRANSLATE) built into SQL
ML Functions in Cortex: FORECAST, ANOMALY_DETECTION, CLASSIFICATION — no-code ML
Cortex runs serverlessly on Snowflake GPU infrastructure

Quick Feedback

Snowflake SnowPro Core (COF-C03) Study Notes

Domain Weight

Three-Layer Architecture

Micro-Partitions

Virtual Warehouses

Snowflake Editions

Snowpark & Cortex AI

Snowflake SnowPro Core (COF-C03) Study Notes

Domain Weight

Three-Layer Architecture

Micro-Partitions

Virtual Warehouses

Snowflake Editions

Snowpark & Cortex AI