ClickHouse OLAP AI Skill Guide

Overview & Engine Architecture

ClickHouse is a columnar OLAP DBMS: data lives in parts merged in the background; ORDER BY (primary key order) enables sparse indexing—not a traditional PK uniqueness guarantee. PARTITION BY prunes scans; tiny inserts create too many parts (merge storms). Replication uses ZooKeeper/ClickHouse Keeper; Kafka engine tables buffer ingest.

Batch inserts / Kafka engine / S3 insert
    -> MergeTree parts on disk
        -> background merges
        -> SELECT with partition + primary-key pruning
        -> optional projections / materialized views

When to use / when not to

Use when