Module 21 · Core

How engines read a lakehouse

Follow one SQL query from catalog lookup through pruning to vectorised execution, across Spark, Trino, DuckDB and more.

In plain words

When you run SQL on a lakehouse, an engine looks up the table, reads its metadata, skips every file it can, and processes the rest column by column. Following one query end to end shows why the whole stack is designed the way it is.

Helps to take first

Inside a Parquet file What makes a table a table

Words you'll meet

Query engine Query plan Predicate pushdown Data skipping Min/max statistics

This module is in production

Here's what it will cover.

Centrepiece

The life of a SQL query: scroll from SQL text to a plan to pruned files to results, then compare how Spark, Trino and DuckDB do it

You will understand

Planning: catalog → metadata → file pruning → scan
Vectorised, columnar execution
Engines: Spark, Trino/Presto, Flink, DuckDB, StarRocks, Dremio
Serverless engines: Athena, BigQuery, Snowflake, Databricks SQL

Formats

Scroll storyStep-throughCheckpoints