Apache Iceberg: Multi-Engine Lakehouse, Spec to Production

Apache Iceberg won the table-format war. Now you have to actually run one in production.

Most tutorials show you a CREATE TABLE and call it a lakehouse. This course takes you all the way: the spec internals, every major catalog, all the write and read engines, and the production discipline that separates a weekend pilot from a platform 200 engineers depend on.

Across 23 modules and 116 lessons (~12.5 hours), you'll follow Sarah — a Head of Data Platform running shared Iceberg storage across three continents — as she makes the decisions you'll soon make yourself. You'll read the spec as a working mental model (manifests, snapshots, metadata.json, the commit protocol), then choose a catalog with a real Decision Spine matrix comparing Polaris, Unity OSS, Nessie, Glue, and Hive Metastore — and defend the choice.

From there you build: a complete medallion lakehouse with Kafka CDC into bronze, Spark MERGE plus write-audit-publish into silver, and multi-engine readers (Trino, Snowflake, DuckDB) on gold. You'll run the maintenance nobody teaches — compaction, expire-snapshots, orphan cleanup, rewrite-manifests — with real cost numbers, and master branching, tagging, and time-travel rollback patterns that auditors and SREs sign off on.

What makes this course different

• Spec-to-production, not hello-world. You finish able to operate Iceberg under real load, not just create a table.
• Decision-first. A six-version Decision Spine (format, catalog, storage cost, engine cost, migration, an 8-dimension capstone rubric) gives you a defensible framework, not vendor hype.
• Genuinely multi-engine. Spark, Flink, Kafka Connect, Trino, Snowflake, DuckDB — one storage, many engines, the way real lakehouses are built.
• War stories + a 10-step diagnosis playbook for the four failures every Iceberg team eventually hits.

You'll learn to

• Read the Iceberg spec — manifests, snapshots, metadata.json, commit protocol — as a mental model.
• Pick the right catalog with the Decision Spine matrix and defend it.
• Ship a medallion lakehouse with Kafka CDC, Spark MERGE, and multi-engine reads.
• Run production maintenance (compaction, expire-snapshots, orphan-cleanup, rewrite-manifests) with cost numbers.
• Apply WAP, branching, tagging, and time-travel rollback patterns.
• Migrate Hive, Delta, or bare Parquet to Iceberg with the right strategy.
• Diagnose the four most common Iceberg failures with a 10-step playbook.

Who's teaching

Built by Snowbrix Academy and taught by Amit — a credentialed practitioner (SnowPro Core, 2x Databricks-certified) who builds production data platforms for a living. Every pattern here is one you can defend on Monday.

If you're a data engineer ready to own the lakehouse — to stop following table-format tutorials and start architecting and operating one — this is your course.