Snowflake Mastery Path¶
Snowflake is easy to start with: create a warehouse, load a file, run SQL. It's also easy to get surprising results and
surprising bills. A load that quietly keeps 199,997 of 200,000 rows, a FLATTEN that drops a quarter of your customers, a
one-day query that reads every partition because of how rows arrived, a join that makes revenue six times too high, a
MERGE that inserts the same order twice, administrators who can read raw PII through an innocent role grant, a warehouse
bill that's 86% idle time, and a dbt model that passes every test while serving stale numbers: none of these are Snowflake
bugs. They follow from how Snowflake actually works, and this course shows each one.
It assumes you know SQL and some Python. It doesn't assume any Snowflake experience.
Level 1 covers the foundations: the three-layer architecture and credits, accounts and clients (Snowflake CLI and
Python with key pairs), objects and types, warehouses and billing, loading with stages and COPY, Snowflake SQL,
semi-structured JSON, Time Travel and Fail-safe, and zero-copy cloning, ending with a small analytics database.
Level 2 is performance and data engineering: micro-partitions and pruning, clustering, the Query Profile, the three
caches, Snowpipe, streams, tasks, dynamic tables, unloading and Iceberg, ending with an incremental ELT pipeline.
Level 3 is security and governance: roles and privileges, authentication after the end of single-factor passwords,
network policies, masking and row access policies, tags, secure views, data sharing, encryption and auditing, ending with
a governed multi-team platform. Level 4 is building and operating: UDFs and procedures, Snowpark, Python UDFs, Cortex
AI functions, cost management, warehouse scaling, replication and failover, dbt and CI/CD, and migration, ending with a
cost-aware platform whose chargeback is tested.
Every lesson has a How It Actually Works section: why a filter on one column can skip 97 of 98 micro-partitions and the same filter on another can't skip any, why a clone costs nothing until it diverges, why a secure view is slower, why a masking rule that names one role also unmasks every role above it, and why per-query cost attribution never adds up to the bill.
Versions and what was actually run
This course was written without access to a Snowflake account, and no output is presented as Snowflake output. Every Snowflake SQL statement was parsed with sqlglot 30.22 (Snowflake dialect) and checked against Snowflake's documentation as of October 2026. Where a result is shown, it comes from something that was really run, and is labelled: the same SQL on the same generated data in DuckDB 1.5.6, Python models of the documented billing and replication rules (with their assumptions stated), Snowflake CLI 3.29.0 and the Python connector for everything that doesn't contact Snowflake, Snowpark 1.53.0 in its local testing mode, dbt-core 1.12.5, and pytest suites for every project. Things that couldn't be run, such as Cortex AI output, real query profiles and failover, are described from the documentation and never shown with invented results.
Snowflake changes quickly, and several features here are new in 2025–2026: mandatory MFA and the end of single-factor passwords, programmatic access tokens, Snowpipe's per-GB pricing, task overlap policies, Cortex AI functions and model access control, Gen2 as the default warehouse, Adaptive Warehouses, Optimized Refresh and RPO Assurance for replication, and dbt Projects on Snowflake. Each lesson says what it was checked against, so you can compare with your account.
How the program is organized¶
| Level | Focus | Modules | What you can do after |
|---|---|---|---|
| 1 · Snowflake Foundations | Architecture, clients, objects, warehouses, loading, SQL, JSON, Time Travel, cloning | 10 | Build and query a well-organized database and predict its cost |
| 2 · Performance & Data Engineering | Pruning, clustering, profiles, caches, Snowpipe, streams, tasks, dynamic tables, Iceberg | 10 | Make queries read less and keep data fresh incrementally |
| 3 · Security, Governance & Sharing | RBAC, authentication, network, masking, tags, secure views, sharing, encryption, auditing | 10 | Run a shared platform where access is designed, tested and audited |
| 4 · Building, Scaling & Operating | UDFs, Snowpark, Cortex AI, cost, scaling, replication, dbt CI/CD, migration | 10 | Build on Snowflake and keep it fast, recoverable and affordable |
Each level has 9 teaching modules plus a hands-on project (module 10), for 40 lessons total.
How to use this site¶
- Use a free trial account. Every exercise fits in Snowflake's free trial. Use your own account, never a shared or
production one, and give every warehouse
AUTO_SUSPEND = 60from the start. - Run the examples twice. Once as written, and once on your account. Results shown from DuckDB or a model are labelled, and comparing them with what Snowflake does is part of the learning.
- Watch the meter. Keep an eye on
WAREHOUSE_METERING_HISTORYwhile you work. The cost lessons are easier when you've seen your own numbers. - Do the projects. Each one combines its level into something you could adapt for real work, and each comes with tests, including the bugs those tests caught.
More from the Mastery Path series¶
Free, structured, module-wise training across 86 other languages, platforms and disciplines:
Languages
Web Frameworks
Mobile
Testing & QA
Security
Cloud Platforms
Data & Analytics
Databases
AI / ML / LLM
Embedded Systems
Leadership & Management
Professional Skills
Careers & Interviews
Process & APIs
Infrastructure & Ops