An Overview of ClickHouse: Features, Architecture, and Benchmarks

ClickHouse is a column-oriented database management system (DBMS) for online analytical processing of queries (OLAP).

ClickHouse, short for Click Stream Data Warehouse, originated from Yandex’s analytics platform and is an open‑source analytical database developed by the Russian company Yandex.

Yandex uses ClickHouse in its Yandex.Metrica service to analyze massive web traffic, similar to Google Analytics or Baidu Tongji.

According to the official benchmark, ClickHouse outperforms several other databases on identical hardware, achieving up to 429× faster response than MySQL and 126× faster than Hive.

Key Features of ClickHouse

ROLAP (relational online analytical processing) versus OLTP.

Real‑time online queries.

Full DBMS capabilities.

Pure columnar storage (different from column‑family stores like HBase).

No need for data pre‑processing.

Supports batch updates.

Comprehensive SQL support and functions.

High availability with multi‑master architecture.

Operates out‑of‑the‑box without a Hadoop ecosystem.

Some Limitations

Does not support transactions (common for OLAP databases).

Not optimized for row‑level primary‑key lookups (though possible).

Row‑level deletions are not efficient (though supported).

ClickHouse Architecture

1. Column and Field

Columns represent whole columns in memory via the IColumn interface; Fields represent individual scalar values within a column.

2. DataType

DataType handles serialization/deserialization and delegates actual data reading to Column or Field objects.

3. Block and Block Streams

Data processing is performed on Block objects, which contain columns, data types, and column names. IBlockInputStream reads data, while IBlockOutputStream writes data; implementations cover DDL, relational operations, and table‑engine specific processing.

4. Table (IStorage)

Tables are represented by the IStorage interface; different table engines implement this interface to define, query, and write data.

5. Parser and Interpreter

The Parser creates an AST from SQL; the Interpreter walks the AST, builds the execution pipeline, and returns IBlock results.

6. Functions and Aggregate Functions

Functions (IFunction) operate vectorially on whole columns; Aggregate Functions (IAggregateFunction) maintain state (e.g., COUNT) and can be serialized across nodes.

7. Cluster and Replication

Clusters consist of shards, each with replicas; a node hosts a single shard, and replication provides fault tolerance.

ClickHouse Table Engines

MergeTree – primary‑key and date‑based indexing with real‑time updates.

ReplacingMergeTree – deduplicates rows with the same primary key during merges.

SummingMergeTree – aggregates rows with identical keys, useful for summarizing numeric columns.

AggregatingMergeTree – stores aggregate function states for incremental materialized view calculations.

Distributed – performs distributed queries across multiple servers without storing data locally.