Percona Binary Log Server

binlog_server is a command line utility that can be considered as an enhanced version of mysqlbinlog in --read-from-remote-server mode which serves as a replication client and can stream binary log events from a remote Oracle MySQL Server / Percona Server for MySQL both to a local filesystem and to a cloud storage (currently to AWS S3). It is capable of automatically reconnecting to the remote server and resume operation from the point when it was previously stopped / terminated.

It is written in portable c++ following the c++20 standard best practices.

Installation

Prebuilt Binaries

Currently prebuilt binaries are not available.

Building from source

Dependencies

• CMake 3.20.0+
• Clang (clang-15 / clang-16 / clang-17) or GCC (gcc-12 / gcc-13 / gcc-14)
• Boost libraries 1.84.0 (git version, not the source tarball)
• MySQL client library 8.0.x (libmysqlclient)
• CURL library (libcurl) 8.6.0+
• AWS SDK for C++ 1.11.286

Instructions

Creating a build workspace

mkdir ws

Every next step will assume that we are currently inside the ws directory unless explicitly stated otherwise.

Defining environment variables affecting build configurations

Define BUILD_CONFIGURATION depending on whether you want to build Debug or Release version of the library.

export BUILD_CONFIGURATION=Debug

or

export BUILD_CONFIGURATION=RelWithDebInfo

Define BUILD_CC and BUILD_CXX depending on the compiler you wish to use.

export BUILD_CC=gcc
export BUILD_CXX=g++

Instead of gcc you may use gcc-12 / gcc-13 / gcc-14 or clang-15 / clang-16 / clang-17. Instead of g++ you may use g++-12 / g++-13 / g++-14 or clang++-15 / clang++-16 / clang++-17. Make sure that versions of the CC and CXX compilers match.

Boost Libraries

Getting Boost Libraries source

git clone --recurse-submodules --jobs=8 https://github.com/boostorg/boost.git
cd boost
git checkout --recurse-submodules -b required_release boost-1.84.0

Configuring Boost Libraries

cmake \
  -B ./boost-build-${BUILD_CONFIGURATION} \
  -S ./boost \
  -DCMAKE_INSTALL_PREFIX=./boost-install-${BUILD_CONFIGURATION} \
  -DCMAKE_BUILD_TYPE=${BUILD_CONFIGURATION} \
  -DCMAKE_C_COMPILER=${BUILD_CC} \
  -DCMAKE_CXX_COMPILER=${BUILD_CXX} \
  -DBUILD_SHARED_LIBS=OFF \
  -DBUILD_TESTING=OFF

Building Boost Libraries

cmake --build ./boost-build-${BUILD_CONFIGURATION} --config ${BUILD_CONFIGURATION} --parallel

Installing Boost Libraries

cmake --install ./boost-build-${BUILD_CONFIGURATION} --config ${BUILD_CONFIGURATION}

AWS SDK CPP Libraries

Getting AWS SDK CPP Libraries source

git clone --recurse-submodules --jobs=8 https://github.com/aws/aws-sdk-cpp
cd aws-sdk-cpp
git checkout --recurse-submodules -b required_release 1.11.286

Configuring AWS SDK CPP Libraries

cmake \
  -B ./aws-sdk-cpp-build-${BUILD_CONFIGURATION} \
  -S ./aws-sdk-cpp \
  -DCMAKE_INSTALL_PREFIX=./aws-sdk-cpp-install-${BUILD_CONFIGURATION} \
  -DCMAKE_BUILD_TYPE=${BUILD_CONFIGURATION} \
  -DCMAKE_C_COMPILER=${BUILD_CC} \
  -DCMAKE_CXX_COMPILER=${BUILD_CXX} \
  -DCPP_STANDARD=20 \
  -DENABLE_UNITY_BUILD=ON \
  -DBUILD_SHARED_LIBS=OFF \
  -DFORCE_SHARED_CRT=OFF \
  -DENABLE_TESTING=OFF \
  -DAUTORUN_UNIT_TESTS=OFF \
  -DBUILD_ONLY=s3-crt

Building AWS SDK CPP Libraries

cmake --build ./aws-sdk-cpp-build-${BUILD_CONFIGURATION} --config ${BUILD_CONFIGURATION} --parallel

Installing AWS SDK CPP Libraries

cmake --install ./aws-sdk-cpp-build-${BUILD_CONFIGURATION} --config ${BUILD_CONFIGURATION}

Main Application

Getting Main Application source

git clone https://github.com/Percona-Lab/percona-binlog-server.git

Configuring Main Application

cmake  \
  -B ./percona-binlog-server-build-${BUILD_CONFIGURATION} \
  -S ./percona-binlog-server \
  -DCMAKE_BUILD_TYPE=${BUILD_CONFIGURATION} \
  -DCMAKE_C_COMPILER=${BUILD_CC} \
  -DCMAKE_CXX_COMPILER=${BUILD_CXX} \
  -DCPP_STANDARD=20 \
  -DCMAKE_PREFIX_PATH=${PWD}/aws-sdk-cpp-install-${BUILD_CONFIGURATION} \
  -DBoost_ROOT=${PWD}/boost-install-${BUILD_CONFIGURATION}

Building Main Application

cmake --build ./percona-binlog-server-build-${BUILD_CONFIGURATION} --config ${BUILD_CONFIGURATION} --parallel

Main Application binary

The result binary can be found under the following path ws/percona-binlog-server-build-${BUILD_CONFIGURATION}/binlog_server

Usage

Command line arguments

Please run

./binlog_server <operation_mode> <json_config_file>`

where <operation_mode> can be either fetch or pull and <json_config_file> is a path to a JSON configuration file (described below).

Operation modes

Percona Binary Log Server utility can operate in two modes:

• 'fetch'
• 'pull'

'fetch' operation mode

In this mode the utility tries to connect to a remote MySQL server, switch connection to replication mode and read events from all available binary logs already stored on the server. After reading the very last event, the utility gracefully disconnects and exits. Any error (network issues, server down, out of space, etc) encountered in this mode results in immediate termination of the program making sure that storage is left in consistent state.

'pull' operation mode

In this mode the utility continuously tries to connect to a remote MySQL server / switch to replication mode and read binary log events. After reading the very last one, the utility does not close the connection immediately but instead waits for <connection.read_timeout> seconds for the server to generate more events. If this period of time elapses, the utility closes the MySQL connection and enters the idle mode. In this mode it just waits for <replication.idle_time> seconds in disconnected state. After that another reconnection attempt is made and everything starts from the beginning. Any network-related error (network issues, server down, etc) encountered in this mode does not result in immediate termination of the program. Instead, another reconnection attempt is made. More serious errors (like out of space, etc.) cause program termination.

JSON Configuration file

The Percona Binary Log Server configuration file has the following format.

{
  "logger": {
    "level": "debug",
    "file": "binsrv.log"
  },
  "connection": {
    "host": "127.0.0.1",
    "port": 3306,
    "user": "rpl_user",
    "password": "rpl_password",
    "connect_timeout": 20,
    "read_timeout": 60,
    "write_timeout": 60
  },
  "replication": {
    "server_id": 42,
    "idle_time": 10
  },
  "storage": {
    "uri": "file:///home/user/vault"
  }
}

<logger> section

• <logger.level> sets the minimum severity of the log messages that user want to appear in the log output, can be one of the trace / debug / info / warning / error / fatal (explained below).
• <logger.file> can be either a path to a file on a local filesytem to which all log messages will be written or an empty string "" meaning that all the output will be made to console (STDOUT).

Logger message severity levels

Each message written to the log has the severity level associated with it. Currently we use the following mapping:

• fatal - currently not used,
• error - used for printing messages coming from caught exceptions,
• warning - currently not used,
• info - primary log severity level used mostly to indicate progress (configuration file read, storage created, connection established, etc.),
• debug - used to print function names from caught exceptions and to print the data from parsed binary log events,
• trace - used to print source file name / line number / position from caught exceptions and to print raw data (hex dumps) of binary log events.

<connection> section

• <connection.host> - MySQL server host name (e.g. 127.0.0.1, 192.168.0.100, dbsrv.mydomain.com, etc.). Please do not use localhost here as it will be interpreted differently by the libmysqlclient and will instruct the library to use Unix socket file for connection instead of TCP protocol - use 127.0.0.1 instead (see this page for more details).
• <connection.port> - MySQL server port (e.g. 3306 - the default MySQL server port).
• <connection.user> - the name of the MySQL user that has REPLICATION SLAVE privilege.
• <connection.password> - the password for this MySQL user
• <connection.connect_timeout> - the number of seconds the MySQL client library will wait to establish a connection with a remote host.
• <connection.read_timeout> - the number of seconds the MySQL client library will wait to read data from a remote server (this parameter may affect the responsiveness of the program to graceful termination - see below).
• <connection.write_timeout> - the number of seconds the MySQL client library will wait to write data to a remote server.

<replication> section

• <replication.server_id> - specifies the server ID that the utility will be using when connecting to a remote MySQL server (similar to --connection-server-id mysqlbinlog command line option).
• <replication.idle_time> - the number of seconds the utility will spend in disconnected mode between reconnection attempts.

<storage> section

• <storage.uri> - specifies the location (either local or remote) where the received binary logs should be stored

Storage URI format

The type of the storage where the received binary logs should be stored (currently, on a local filesystem of on AWS S3) is determined by the protocol of the URI.

• file://... - for local filesystem.
• s3://... - for AWS S3.

Local filesystem storage URIs

In case of local filesystem, the URIs must have the following format. file://<local_fs_path>, where <local_fs_path> is an absolute path on a local filesystem to a directory where downloaded binary log files must be stored. Relative paths are not supported. For instance, file:///home/user/vault.

Please notice 3 forward slashes / (2 from the protocol part file:// and 1 from the absolute path).

AWS S3 storage URIs

In case of AWS S3, the URIs must have the following format. s3://[<access_key_id>:<secret_access_key>@]<bucket_name>[.<region>]/<path>, where:

• <access_key_id> - the AWS key ID (the <access_key_id> / <secret_access_key> pair is optional),
• <secret_access_key>- the AWS secret access key (the <access_key_id> / <secret_access_key> pair is optional),
• <bucket_name> - the name of AWS S3 bucket in which the data must be stored,
• <region> - the name of the AWS region (e.g. us-east-1) where this bucket was created (optional, if omitted, it will be auto-detected),
• <path> - a virtual path (key prefix) inside the bucket under which all the binary log files will be stored.

For example:

• s3://binsrv-bucket/vault - no AWS credentials specified, binsrv-bucket bucket must be publicly write-accessible, the region will be auto-detected, /vault will be the virtual directory.
• s3://binsrv-bucket.us-east-1/vault - no AWS credentials specified, binsrv-bucket bucket must be publicly write-accessible, the bucket must be created in the us-east-1 region, /vault will be the virtual directory.
• s3://key_id:[email protected]/vault - key_id will be used as AWS_ACCESS_KEY_ID, secret will be used as AWS_SECRET_ACCESS_KEY, binsrv-bucket will be the name of the bucket, the bucket must be created in the us-east-1 region, the bucket may have restricted write access, /vault will be the virtual directory.

Resuming previous operation

Running the utility for the second time (in any mode) results in resuming streaming from the position at which the previous run finished.

Graceful termination

The user can request the utility operating in either fetch or pull mode to be gracefully terminated leaving storage in consistent state. For this, the utility sets custom handlers for the the following POSIX signals.

• SIGINT - for processing ^C in console.
• SIGTERM - for processing kill <pid>.

Because of the synchronous nature of the binlog API from the MySQL client library, there still may be a delay between receiving the signal and reacting to it. Worst case scenario, user will have to wait for <connection.read_timeout> seconds (the value from the configuration) + 1 (the granularity of sleep intervals in the idle mode) seconds.

Please note that killing the program with kill -9 <pid> does not guarantee to flush all the internal file buffers / upload temporary data to a cloud storage and may result in losing some progress.

Licensing

Percona is dedicated to keeping open source open. Whenever possible, we strive to include permissive licensing for both our software and documentation. For this project, we are using version 2 of the GNU General Public License (GPLv2).