mysql-ch-replicator

mysql_ch_replicator

mysql_ch_replicator is a powerful and efficient tool designed for real-time replication of MySQL databases to ClickHouse.

With a focus on high performance, it utilizes batching heavily and uses C++ extension for faster execution. This tool ensures seamless data integration with support for migrations, schema changes, and correct data management.

Table of Contents

• Features
• Installation
  • Requirements
  • Installation
  • Docker Installation
• Usage
  • Basic Usage
  • One Time Data Copy
  • Configuration
    • Required settings
    • Optional settings
  • Advanced Features
    • Migrations & Schema Changes
    • Recovery Without Downtime
    • Known Limitations
• Production usage
• Development
  • Running Tests
• Contribution
• License
• Acknowledgements

Features

• Real-Time Replication: Keeps your ClickHouse database in sync with MySQL in real-time.
• High Performance: Utilizes batching and ports slow parts to C++ (e.g., MySQL internal JSON parsing) for optimal performance (±20K events / second on a single core).
• Supports Migrations/Schema Changes: Handles adding, altering, and removing tables without breaking the replication process (*for most cases, details here).
• Recovery without Downtime: Allows for preserving old data while performing initial replication, ensuring continuous operation.
• Correct Data Removal: Unlike MaterializedMySQL, mysql_ch_replicator ensures physical removal of data.
• Comprehensive Data Type Support: Accurately replicates most data types, including JSON, booleans, and more. Easily extensible for additional data types.
• Multi-Database Handling: Replicates the binary log once for all databases, optimizing the process compared to MaterializedMySQL, which replicates the log separately for each database.

Installation

Requirements

• Linux / MacOS
• python3.10 or higher

Installation

To install mysql_ch_replicator, use the following command:

pip install --upgrade mysql_ch_replicator

You may need to also compile C++ components if they're not pre-built for your platform.

Docker Installation

Alternatively, you can use the pre-built Docker image from DockerHub:

docker pull fippo/mysql-ch-replicator:latest

To run the container:

docker run -d \
  -v /path/to/your/config.yaml:/app/config.yaml \
  -v /path/to/your/data:/app/data \
  fippo/mysql-ch-replicator:latest \
  --config /app/config.yaml run_all

Or with environment variables for credentials:

docker run -d \
  -v /path/to/your/config.yaml:/app/config.yaml \
  -v /path/to/your/data:/app/data \
  -e MYSQL_USER=root \
  -e MYSQL_PASSWORD=secret \
  -e CLICKHOUSE_USER=default \
  -e CLICKHOUSE_PASSWORD=secret \
  fippo/mysql-ch-replicator:latest \
  --config /app/config.yaml run_all

Make sure to:

• Mount your configuration file using the -v flag
• Mount a persistent volume for the data directory
• Adjust the paths according to your setup
• Optionally use -e flags to override credentials via environment variables

Usage

Basic Usage

Here is a video tutorial

For realtime data sync from MySQL to ClickHouse:

• Prepare config file. Use example_config.yaml as an example.
• Configure MySQL and ClickHouse servers:

• MySQL server configuration file my.cnf should include following settings (required to write binary log in raw format, and enable password authentication):

🛠 MySQL Config

[mysqld]
# ... other settings ...
gtid_mode = on
enforce_gtid_consistency = 1
binlog_expire_logs_seconds = 864000
max_binlog_size            = 500M
binlog_format              = ROW

• For MariaDB use following settings:

[mysqld]
# ... other settings ...
gtid_strict_mode = ON
gtid_domain_id = 0
server_id = 1
log_bin = /var/log/mysql/mysql-bin.log
binlog_expire_logs_seconds = 864000
max_binlog_size = 500M
binlog_format = ROW

For AWS RDS you need to set following settings in Parameter groups:

binlog_format                       ROW
binlog_expire_logs_seconds          86400

• ClickHouse server config override.xml should include following settings (it makes clickhouse apply final keyword automatically to handle updates correctly):

🛠 ClickHouse Config

<clickhouse>
    <!-- ... other settings ... -->
    <profiles>
        <default>
            <!-- ... other settings ... -->
            <final>1</final>
            <max_query_size>300000000</max_query_size>
            <max_ast_elements>1000000</max_ast_elements>
            <max_expanded_ast_elements>1000000</max_expanded_ast_elements>
        </default>
    </profiles>
</clickhouse>

!!! Double check final setting is applied !!!

Execute the following command in clickhouse:

SELECT name, value, changed FROM system.settings WHERE name = 'final' Setting should be set to 1. If not, you should:

• double check the override.xml is applied
• try to modify users.xml instead

• Start the replication:

mysql_ch_replicator --config config.yaml run_all

This will keep data in ClickHouse updating as you update data in MySQL. It will always be in sync.

One Time Data Copy

If you just need to copy data once, and don't need continuous synchronization for all changes, you should do following:

• Prepare config file. Use example_config.yaml as an example.
• Run one-time data copy:

mysql_ch_replicator --config config.yaml db_replicator --db mysql_db_name --initial_only=True

Where mysql_db_name is the name of the database you want to copy.

Don't be afraid to interrupt process in the middle. It will save the state and continue copy after restart.

Hint: set initial_replication_threads to a number of cpu cores to accelerate initial replication

Configuration

mysql_ch_replicator can be configured through a configuration file and also by using enviromnent variables to override some of config settings. Here is the config example:

mysql:
  host: 'localhost'
  port: 8306
  user: 'root'
  password: 'root'
  charset: 'utf8mb4'  # optional, default is utf8mb4 for full Unicode support

clickhouse:
  host: 'localhost'
  port: 8323
  user: 'default'
  password: 'default'
  connection_timeout: 30        # optional
  send_receive_timeout: 300     # optional

binlog_replicator:
  data_dir: '/home/user/binlog/'  # a new EMPTY directory (for internal storage of data by mysql_ch_replicator itself)
  records_per_file: 100000
  binlog_retention_period: 43200  # optional, how long to keep binlog files in seconds, default 12 hours

databases: 'database_name_pattern_*'
tables: '*'


# OPTIONAL SETTINGS

initial_replication_threads: 4                        # optional

exclude_databases: ['database_10', 'database_*_42']   # optional
exclude_tables: ['meta_table_*']                      # optional

target_databases:                                     # optional
  source_db_in_mysql_1: destination_db_in_clickhouse_1
  source_db_in_mysql_2: destination_db_in_clickhouse_2
  ...

target_tables:                                        # optional
  source_db_in_mysql_1.source_table_in_mysql_1: destination_table_name_1
  source_db_in_mysql_1.source_table_in_mysql_2: destination_table_name_2
  ...

log_level: 'info'               # optional       
optimize_interval: 86400        # optional
auto_restart_interval: 3600     # optional

indexes:                        # optional
  - databases: '*'
    tables: ['test_table']
    index: 'INDEX name_idx name TYPE ngrambf_v1(5, 65536, 4, 0) GRANULARITY 1'

partition_bys:                  # optional
  - databases: '*'
    tables: ['test_table']
    partition_by: 'toYYYYMM(created_at)'

http_host: '0.0.0.0'    # optional
http_port: 9128         # optional

types_mapping:          # optional
  'char(36)': 'UUID'

ignore_deletes: false    # optional, set to true to ignore DELETE operations

mysql_timezone: 'UTC'    # optional, timezone for MySQL timestamp conversion (default: 'UTC')

Required settings

• mysql MySQL connection settings
• clickhouse ClickHouse connection settings
• binlog_replicator.data_dir Create a new empty directory, it will be used by script to store it's state
• databases Databases name pattern to replicate, e.g. db_* will match db_1 db_2 db_test, list is also supported

Optional settings

• initial_replication_threads - number of threads for initial replication, by default 1, set it to number of cores to accelerate initial data copy
• tables - tables to filter, list is also supported
• exclude_databases - databases to exclude, string or list, eg 'table1*' or ['table2', 'table3*']. If same database matches databases and exclude_databases, exclude has higher priority.
• exclude_tables - databases to exclude, string or list. If same table matches tables and exclude_tables, exclude has higher priority.
• target_databases - if you want database in ClickHouse to have different name from MySQL database
• target_tables - if you want table in ClickHouse to have different name from MySQL table. Specify as source_database.source_table: target_table_name. The target database is determined by existing rules (e.g., target_databases mapping). This mapping applies to both initial and realtime replication, including DDL operations like ALTER, DROP, etc.
• log_level - log level, default is info, you can set to debug to get maximum information (allowed values are debug, info, warning, error, critical)
• optimize_interval - interval (seconds) between automatic OPTIMIZE table FINAL calls. Default 86400 (1 day). This is required to perform all merges guaranteed and avoid increasing of used storage and decreasing performance.
• auto_restart_interval - interval (seconds) between automatic db_replicator restart. Default 3600 (1 hour). This is done to reduce memory usage.
• binlog_retention_period - how long to keep binlog files in seconds. Default 43200 (12 hours). This setting controls how long the local binlog files are retained before being automatically cleaned up.
• indexes - you may want to add some indexes to accelerate performance, eg. ngram index for full-test search, etc. To apply indexes you need to start replication from scratch.
• partition_bys - custom PARTITION BY expressions for tables. By default uses intDiv(id, 4294967) for integer primary keys. Useful for time-based partitioning like toYYYYMM(created_at).
• http_host, http_port - http endpoint to control replication, use /docs for abailable commands
• types_mappings - custom types mapping, eg. you can map char(36) to UUID instead of String, etc.
• ignore_deletes - when set to true, DELETE operations in MySQL will be ignored during replication. This creates an append-only model where data is only added, never removed. In this mode, the replicator doesn't create a temporary database and instead replicates directly to the target database.
• mysql_timezone - timezone to use for MySQL timestamp conversion to ClickHouse DateTime64. Default is 'UTC'. Accepts any valid timezone name (e.g., 'America/New_York', 'Europe/London', 'Asia/Tokyo'). This setting ensures proper timezone handling when converting MySQL timestamp fields to ClickHouse DateTime64 with timezone information.
• post_initial_replication_commands - SQL commands to execute in ClickHouse after initial replication completes for each database. Useful for creating materialized views, summary tables, or other database objects. Commands are executed in order, once per database matching the pattern.

post_initial_replication_commands:
  - databases: '*'
    commands:
      - 'CREATE TABLE IF NOT EXISTS summary_table (...) ENGINE = SummingMergeTree() ORDER BY (...)'
      - 'CREATE MATERIALIZED VIEW IF NOT EXISTS data_mv TO summary_table AS SELECT ...'
      - 'INSERT INTO summary_table SELECT ... FROM replicated_table'

Few more tables / dbs examples:

databases: ['my_database_1', 'my_database_2']
tables: ['table_1', 'table_2*']

Environment Variables: MySQL and ClickHouse credentials can be overridden using environment variables for better security in containerized environments:

• MySQL: MYSQL_HOST, MYSQL_PORT, MYSQL_USER, MYSQL_PASSWORD, MYSQL_CHARSET
• ClickHouse: CLICKHOUSE_HOST, CLICKHOUSE_PORT, CLICKHOUSE_USER, CLICKHOUSE_PASSWORD

Advanced Features

Migrations & Schema Changes

mysql_ch_replicator supports the following:

• Adding Tables: Automatically starts replicating data from newly added tables.
• Altering Tables: Adjusts replication strategy based on schema changes.
• Removing Tables: Handles removal of tables without disrupting the replication process.

WARNING. While 95% of operations supported, there could be still some unhandled operations. We try to support all of them, but for your safety, please write the CI/CD test that will check your migrations. Test should work a following way:

• Aplly all your mysql migrations
• Try to insert some record into mysql (to any table)
• Check that this record appears in ClickHouse

Known Limitations

• Migrations not supported during initial replication. You should either wait for initial replication finish and then apply migrations, or restart initial replication from scratch (by removing state file).
• Primary key changes not supported. This is a ClickHouse level limitation, it does not allow to make any changes realted to primary key.
• Tables without a primary key are not supported.

Recovery Without Downtime

In case of a failure or during the initial replication, mysql_ch_replicator will preserve old data and continue syncing new data seamlessly. You could remove the state and restart replication from scratch.

Production usage

Here is the checklist for production. Please read it carefully and confirm every point. If you have any questions - feel free to ask in discussion / issues. Following the checklist is critical to avoid any issues on prod.

Checklist

• No manual changes are made directly to ClickHouse databases or tables. All modifications (indexes, table settings, etc.) must be applied as overrides in the mysql_ch_replicator config.yaml.
• All MySQL migrations undergo either CI/CD or manual testing before deployment. No migration is applied to production without being tested. The tests verify that replication continues to work correctly and that new records are successfully replicated from MySQL to ClickHouse after the migration.
• Read and acknowledged known limitations.

Development

To contribute to mysql_ch_replicator, clone the repository and install the required dependencies:

git clone https://github.com/your-repo/mysql_ch_replicator.git
cd mysql_ch_replicator
pip install -r requirements.txt

Running Tests

• Use docker-compose to install all requirements:

cd tests/
docker compose -f docker-compose-tests.yaml up

• Run tests with:

docker exec -w /app/ -it tests-replicator-1 python3 -m pytest -v -s tests/

• To run a single test:

docker exec -w /app/ -it tests-replicator-1 python3 -m pytest -v -s tests/ -k test_your_test_name

Contribution

Contributions are welcome! Please open an issue or submit a pull request for any bugs or features you would like to add.

License

mysql_ch_replicator is licensed under the MIT License. See the LICENSE file for more details.

Acknowledgements

Thank you to all the contributors who have helped build and improve this tool.