WAL-G

This documentation is also available at wal-g.readthedocs.io

WAL-G is an archival restoration tool for PostgreSQL, MySQL/MariaDB, and MS SQL Server (beta for MongoDB and Redis).

WAL-G is the successor of WAL-E with a number of key differences. WAL-G uses LZ4, LZMA, ZSTD, or Brotli compression, multiple processors, and non-exclusive base backups for Postgres. More information on the original design and implementation of WAL-G can be found on the Citus Data blog post "Introducing WAL-G by Citus: Faster Disaster Recovery for Postgres".

Table of Contents

Installation
Configuration
Usage
Databases
Development
Authors
License
Acknowledgments
Chat

Installation

A precompiled binary for Linux AMD 64 of the latest version of WAL-G can be obtained under the Releases tab.

Binary name has the following format: wal-g-DBNAME-OSNAME, where DBNAME stands for the name of the database (for example pg, mysql), OSNAME stands for the name of the operating system used for building the binary.

To decompress the binary, use:

tar -zxvf wal-g-DBNAME-OSNAME-amd64.tar.gz
mv wal-g-DBNAME-OSNAME-amd64 /usr/local/bin/wal-g

For example, for Postgres and Ubuntu 18.04:

tar -zxvf wal-g-pg-ubuntu-18.04-amd64.tar.gz
mv wal-g-pg-ubuntu-18.04-amd64 /usr/local/bin/wal-g

For other systems, please consult the Development section for more information.

WAL-G supports bash and zsh autocompletion. Run wal-g help completion for more info.

Configuration

Storage

To configure where WAL-G stores backups, please consult the Storages section.

Compression

WALG_COMPRESSION_METHOD

To configure the compression method used for backups. Possible options are: lz4, lzma, zstd, brotli. The default method is lz4. LZ4 is the fastest method, but the compression ratio is bad. LZMA is way much slower. However, it compresses backups about 6 times better than LZ4. Brotli and zstd are a good trade-off between speed and compression ratio, which is about 3 times better than LZ4.

Encryption

YC_CSE_KMS_KEY_ID

To configure Yandex Cloud KMS key for client-side encryption and decryption. By default, no encryption is used.

YC_SERVICE_ACCOUNT_KEY_FILE

To configure the name of a file containing private key of Yandex Cloud Service Account. If not set a token from the metadata service (http://169.254.169.254) will be used to make API calls to Yandex Cloud KMS.

WALG_LIBSODIUM_KEY

To configure encryption and decryption with libsodium. WAL-G uses an algorithm that only requires a secret key. libsodium keys are fixed-size keys of 32 bytes. For optimal cryptographic security, it is recommened to use a random 32 byte key. To generate a random key, you can something like openssl rand -hex 32 (set WALG_LIBSODIUM_KEY_TRANSFORM to hex) or openssl rand -base64 32 (set WALG_LIBSODIUM_KEY_TRANSFORM to base64).

WALG_LIBSODIUM_KEY_PATH

Similar to WALG_LIBSODIUM_KEY, but value is the path to the key on file system. The file content will be trimmed from whitespace characters.

WALG_LIBSODIUM_KEY_TRANSFORM

The transform that will be applied to the WALG_LIBSODIUM_KEY to get the required 32 byte key. Supported transformations are base64, hex or none (default). The option none exists for backwards compatbility, the user input will be converted to 32 byte either via truncation or by zero-padding.

WALG_GPG_KEY_ID (alternative form WALE_GPG_KEY_ID) ⚠️ DEPRECATED

To configure GPG key for encryption and decryption. By default, no encryption is used. Public keyring is cached in the file "/.walg_key_cache".

WALG_PGP_KEY

To configure encryption and decryption with OpenPGP standard. You can join multiline key using \n symbols into one line (mostly used in case of daemontools and envdir). Set private key value when you need to execute wal-fetch or backup-fetch command. Set public key value when you need to execute wal-push or backup-push command. Keep in mind that the private key also contains the public key.

WALG_PGP_KEY_PATH

Similar to WALG_PGP_KEY, but value is the path to the key on file system.

WALG_PGP_KEY_PASSPHRASE

If your private key is encrypted with a passphrase, you should set passphrase for decrypt.

Monitoring

WALG_STATSD_ADDRESS

To enable metrics publishing to statsd or statsd_exporter. Metrics will be sent on a best-effort basis via UDP. The default port for statsd is 9125.

Profiling

Profiling is useful for identifying bottlenecks within WAL-G.

PROFILE_SAMPLING_RATIO

A float value between 0 and 1, defines likelihood of the profiler getting enabled. When set to 1, it will always run. This allows probabilistic sampling of invocations. Since WAL-G processes may get created several times per second (e.g. wal-g wal-push), we do not want to profile all of them.

PROFILE_MODE

The type of pprof profiler to use. Can be one of cpu, mem, mutex, block, threadcreation, trace, goroutine. See the runtime/pprof docs for more information. Defaults to cpu.

PROFILE_PATH

The directory to store profiles in. Defaults to $TMPDIR.

Rate limiting

WALG_NETWORK_RATE_LIMIT

Network traffic rate limit during the backup-push/backup-fetch operations in bytes per second.

Database-specific options

More options are available for the chosen database. See it in Databases

Usage

WAL-G currently supports these commands for all type of databases:

`backup-list`

Lists names and creation time of available backups.

--pretty flag prints list in a table

--json flag prints list in JSON format, pretty-printed if combined with --pretty

--detail flag prints extra backup details, pretty-printed if combined with --pretty, json-encoded if combined with --json

`delete`

Is used to delete backups and WALs before them. By default, delete will perform a dry run. If you want to execute deletion, you have to add --confirm flag at the end of the command. Backups marked as permanent will not be deleted.

delete can operate in four modes: retain, before, everything and target.

retain [FULL|FIND_FULL] %number% [--after %name|time%]

if FULL is specified, keep %number% full backups and everything in the middle. If with --after flag is used keep $number$ the most recent backups and backups made after %name|time% (including).

before [FIND_FULL] %name%

If FIND_FULL is specified, WAL-G will calculate minimum backup needed to keep all deltas alive. If FIND_FULL is not specified, and call can produce orphaned deltas, the call will fail with the list.

everything [FORCE]

target [FIND_FULL] %name% | --target-user-data %data% will delete the backup specified by name or user data. Unlike other delete commands, this command does not delete any archived WALs.

(Only in Postgres) By default, if delta backup is provided as the target, WAL-G will also delete all the dependant delta backups. If FIND_FULL is specified, WAL-G will delete all backups with the same base backup as the target.

Examples

everything all backups will be deleted (if there are no permanent backups)

everything FORCE all backups, include permanent, will be deleted

retain 5 will fail if 5th is delta

retain FULL 5 will keep 5 full backups and all deltas of them

retain FIND_FULL 5 will find necessary full for 5th and keep everything after it

retain 5 --after 2019-12-12T12:12:12 keep 5 most recent backups and backups made after 2019-12-12 12:12:12

before base_000010000123123123 will fail if base_000010000123123123 is delta

before FIND_FULL base_000010000123123123 will keep everything after base of base_000010000123123123

target base_0000000100000000000000C9 delete the base backup and all dependant delta backups

target --target-user-data "{ \"x\": [3], \"y\": 4 }" delete backup specified by user data

target base_0000000100000000000000C9_D_0000000100000000000000C4 delete delta backup and all dependant delta backups

target FIND_FULL base_0000000100000000000000C9_D_0000000100000000000000C4 delete delta backup and all delta backups with the same base backup

More commands are available for the chosen database engine. See it in Databases

Storage tools

wal-g st command series allows the direct interaction with the configured storage. Storage tools documentation

Databases

Development

The following steps describe how to build WAL-G for PostgreSQL, but the process is the same for other databases. For example, to build WAL-G for MySQL, use the make mysql_build instead of make pg_build.

Optional:

To build with brotli compressor and decompressor, set the USE_BROTLI environment variable.
To build with libsodium, set the USE_LIBSODIUM environment variable.
To build with lzo decompressor, set the USE_LZO environment variable.

Ubuntu

# Install latest Go compiler
sudo add-apt-repository ppa:longsleep/golang-backports 
sudo apt update
sudo apt install golang-go

# Install lib dependencies
sudo apt install libbrotli-dev liblzo2-dev libsodium-dev curl cmake

# Fetch project and build
go get github.com/wal-g/wal-g
cd ~/go/src/github.com/wal-g/wal-g
make deps
make pg_build
main/pg/wal-g --version

Users can also install WAL-G by using make pg_install. Specifying the GOBIN environment variable before installing allows the user to specify the installation location. By default, make pg_install puts the compiled binary in the root directory (/).

export USE_BROTLI=1
export USE_LIBSODIUM=1
export USE_LZO=1
make pg_clean
make deps
GOBIN=/usr/local/bin make pg_install

macOS

# brew command is Homebrew for Mac OS
brew install cmake
export USE_BROTLI=1
export USE_LIBSODIUM="true" # since we're linking libsodium later
./link_brotli.sh
./link_libsodium.sh
make install_and_build_pg

To build on ARM64, set the corresponding GOOS/GOARCH environment variables:

env GOOS=darwin GOARCH=arm64 make pg_build

The compiled binary to run is main/pg/wal-g

Testing

WAL-G relies heavily on unit tests. These tests do not require S3 configuration as the upload/download parts are tested using mocked objects. Unit tests can be run using

export USE_BROTLI=1
make unittest

For more information on testing, please consult test, testtools and unittest section in Makefile.

WAL-G will perform a round-trip compression/decompression test that generates a directory for data (e.g., data...), compressed files (e.g., compressed), and extracted files (e.g., extracted). These directories will only get cleaned up if the files in the original data directory match the files in the extracted one.

Test coverage can be obtained using:

export USE_BROTLI=1
make coverage

This command generates coverage.out file and opens HTML representation of the coverage.

Development on Windows

Information about installing and usage

Authors

See also the list of contributors who participated in this project.

License

This project is licensed under the Apache License, Version 2.0, but the lzo support is licensed under GPL 3.0+. Please refer to the LICENSE.md file for more details.

Acknowledgments

WAL-G would not have happened without the support of Citus Data

WAL-G came into existence as a result of the collaboration between a summer engineering intern at Citus, Katie Li, and Daniel Farina, the original author of WAL-E, who currently serves as a principal engineer on the Citus Cloud team. Citus Data also has an open-source extension to Postgres that distributes database queries horizontally to deliver scale and performance.

WAL-G development is supported by Yandex Cloud

Chat

We have a Slack group and Telegram chat to discuss WAL-G usage and development. To join PostgreSQL slack, use invite app.

LeGEC/wal-g