FastFormats: A Benchmark for ClickHouse Input Formats

Overview

FastFormats is a benchmark designed to compare the ingestion performance of different data formats in ClickHouse. It measures server-side insert speed, CPU and memory efficiency, and the impact of compression, pre-sorting, and batching on ingestion performance.

ClickHouse supports 70+ input formats, but choosing the right one can significantly impact performance. FastFormats systematically evaluates these formats to identify the most efficient options for high-throughput ingestion.

Dataset

FastFormats uses the real-world web analytics dataset from ClickBench.

Benchmark Scope

FastFormats evaluates:

• Insert speed: How quickly the server processes data in different formats.
• Resource efficiency: CPU and memory usage per insert.
• Compression impact: How different compression methods affect insert performance.
• Pre-sorting benefits: Whether sorting data before sending improves ingestion speed.
• Batching efficiency: The effect of varying batch sizes on performance.

The benchmark includes 60+ formats, covering:

• Binary columnar formats (e.g., Native, Parquet, Arrow)
• Binary row-oriented formats (e.g., RowBinary, Avro, Protobuf)
• Text-based formats (e.g., CSV, TSV, JSONEachRow)

How It Works

FastFormats runs a nested loop over multiple test dimensions:

• Batch sizes: 10k, 100k, and 1M rows per insert.
• Formats: CSV, TSV, JSONEachRow, Native, Parquet, Arrow, and more.
• Pre-sorting: Yes/No.
• Compression methods: Uncompressed, LZ4, ZSTD.

The benchmark follows these steps:

1. Convert data into different formats using clickhouse-local.
2. Send data to ClickHouse using the HTTP interface.
3. Track ingestion performance, including server-side processing time, CPU, and memory usage.
4. Store results in JSON format for easy analysis.

Blog Post & Results

We wrote a detailed blog post explaining FastFormats, how it works, and showcasing benchmark results for different formats.

📊 For full test results, check out the FastFormats online dashboard: [🔗 Link] (if applicable)

Getting Started

FastFormats requires:

• A client machine to run the benchmark.

• A ClickHouse test system (server) where ClickHouse is running and accessible via HTTP and the native interface.

• curl and clickhouse-client for data ingestion

• clickhouse-local for format conversion

1. Set Up the Client Machine

Clone the Repository

git clone https://github.com/ClickHouse/FastFormats.git
cd FastFormats

Install ClickHouse client tools

FastFormats requires clickhouse-local and clickhouse-client on the client machine.

./install-clickhouse_client.sh

2. Configure ClickHouse Credentials

Set up ClickHouse connection credentials as environment variables:

export CLICKHOUSE_USER="your_user"
export CLICKHOUSE_PASSWORD="your_password"
export CLICKHOUSE_HOST="your_clickhouse_host"

3. Run the Benchmark

After setting up credentials, start the benchmark:

./main.sh

4. Analyze results

Once the benchmark completes, results are stored in the results folder.