docker-volume-backup

Docker image for performing simple backups of Docker volumes. Main features:

• Mount volumes into the container, and they'll get backed up
• Use full cron expressions for scheduling the backups
• Backs up to local disk, to remote host available via scp, to AWS S3, or to all of them
• Allows triggering a backup manually if needed
• Optionally stops containers for the duration of the backup, and starts them again afterward, to ensure consistent backups
• Optionally docker execs commands before/after backing up a container, to allow easy integration with database backup tools, for example
• Optionally executes commands before/after backing up inside docker-volume-backup container and/or on remote host
• Optionally ships backup metrics to InfluxDB, for monitoring
• Optionally encrypts backups with gpg before uploading

Examples

Backing up locally

Say you're running some dashboards with Grafana and want to back them up:

version: "3"

services:

  dashboard:
    image: grafana/grafana:7.4.5
    volumes:
      - grafana-data:/var/lib/grafana           # This is where Grafana keeps its data

  backup:
    image: jareware/docker-volume-backup
    volumes:
      - grafana-data:/backup/grafana-data:ro    # Mount the Grafana data volume (as read-only)
      - ./backups:/archive                      # Mount a local folder as the backup archive

volumes:
  grafana-data:

This will back up the Grafana data volume, once per day, and write it to ./backups with a filename like backup-2018-11-27T16-51-56.tar.gz.

Backing up to S3

Off-site backups are better, though:

version: "3"

services:

  dashboard:
    image: grafana/grafana:7.4.5
    volumes:
      - grafana-data:/var/lib/grafana           # This is where Grafana keeps its data

  backup:
    image: jareware/docker-volume-backup
    environment:
      AWS_S3_BUCKET_NAME: my-backup-bucket      # S3 bucket which you own, and already exists
      AWS_ACCESS_KEY_ID: ${AWS_ACCESS_KEY_ID}   # Read AWS secrets from environment (or a .env file)
      AWS_SECRET_ACCESS_KEY: ${AWS_SECRET_ACCESS_KEY}
    volumes:
      - grafana-data:/backup/grafana-data:ro    # Mount the Grafana data volume (as read-only)

volumes:
  grafana-data:

This configuration will back up to AWS S3 instead. See below for additional tips about S3 Bucket setup.

Restoring from S3

Downloading backups from S3 can be done however you usually interact with S3, e.g. via the aws s3 CLI or the AWS Web Console.

However, if you're on the host that's running this image, you can also download the latest backup with:

$ docker-compose exec -T backup bash -c 'aws s3 cp s3://$AWS_S3_BUCKET_NAME/$BACKUP_FILENAME -' > restore.tar.gz

From here on out the restore process will depend on a variety of things, like whether you've encrypted the backups, how your volumes are configured, and what application it is exactly that you're restoring.

But for the sake of example, to finish the restore for the above Grafana setup, you would:

1. Extract the contents of the backup, with e.g. tar -xf restore.tar.gz. This would leave you with a new directory called backup in the current dir.
2. Figure out the mount point of the grafana-data volume, with e.g. docker volume ls and then docker volume inspect. Let's say it ends up being /var/lib/docker/volumes/bla-bla/_data. This is where your live Grafana keeps its data on the host file system.
3. Stop Grafana, with docker-compose stop dashboard.
4. Move any existing data aside, with e.g. sudo mv /var/lib/docker/volumes/bla-bla/_data{,_replaced_during_restore}. You can also just remove it, if you like to live dangerously.
5. Move the backed up data to where the live Grafana can find it, with e.g. sudo cp -r backup/grafana-data /var/lib/docker/volumes/bla-bla/_data.
6. Depending on the Grafana version, you may need to set some permissions manually, e.g. sudo chown -R 472:472 /var/lib/docker/volumes/bla-bla/_data.
7. Start Grafana back up, with docker-compose start dashboard. Your Grafana instance should now have travelled back in time to its latest backup.

Backing up to remote host by means of SCP

You can also upload to your backups to a remote host by means of secure copy (SCP) based on SSH. To do so, create an SSH key pair if you do not have one yet and copy the public key to the remote host where your backups should be stored. Then, start the backup container by setting the variables SCP_HOST, SCP_USER, SCP_DIRECTORY, and provide the private SSH key by mounting it into /ssh/id_rsa.

In the example, we store the backups in the remote host folder /home/pi/backups and use the default SSH key located at ~/.ssh/id_rsa:

version: "3"

services:

  dashboard:
    image: grafana/grafana:7.4.5
    volumes:
      - grafana-data:/var/lib/grafana           # This is where Grafana keeps its data

  backup:
    image: jareware/docker-volume-backup
    environment:
      SCP_HOST: 192.168.0.42                    # Remote host IP address
      SCP_USER: pi                              # Remote host user to log in
      SCP_DIRECTORY: /home/pi/backups           # Remote host directory
    volumes:
      - grafana-data:/backup/grafana-data:ro    # Mount the Grafana data volume (as read-only)
      - ~/.ssh/id_rsa:/ssh/id_rsa:ro            # Mount the SSH private key (as read-only)

volumes:
  grafana-data:

Triggering a backup manually

Sometimes it's useful to trigger a backup manually, e.g. right before making some big changes.

This is as simple as:

$ docker-compose exec backup ./backup.sh

[INFO] Backup starting

8 containers running on host in total
1 containers marked to be stopped during backup

...
...
...

[INFO] Backup finished

Will wait for next scheduled backup

If you only want to back up manually (i.e. not on a schedule), you should either:

1. Run the image without docker-compose, override the image entrypoint to /root/backup.sh, and ensure you match your env-vars with what the default src/entrypoint.sh would normally set up for you, or
2. Just use BACKUP_CRON_EXPRESSION="#" (to ensure scheduled backup never runs) and execute docker-compose exec backup ./backup.sh whenever you want to run a backup

Stopping containers while backing up

It's not generally safe to read files to which other processes might be writing. You may end up with corrupted copies.

You can give the backup container access to the Docker socket, and label any containers that need to be stopped while the backup runs:

version: "3"

services:

  dashboard:
    image: grafana/grafana:7.4.5
    volumes:
      - grafana-data:/var/lib/grafana           # This is where Grafana keeps its data
    labels:
      # Adding this label means this container should be stopped while it's being backed up:
      - "docker-volume-backup.stop-during-backup=true"

  backup:
    image: jareware/docker-volume-backup
    environment:
      AWS_S3_BUCKET_NAME: my-backup-bucket      # S3 bucket which you own, and already exists
      AWS_ACCESS_KEY_ID: ${AWS_ACCESS_KEY_ID}   # Read AWS secrets from environment (or a .env file)
      AWS_SECRET_ACCESS_KEY: ${AWS_SECRET_ACCESS_KEY}
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro # Allow use of the "stop-during-backup" feature
      - grafana-data:/backup/grafana-data:ro    # Mount the Grafana data volume (as read-only)

volumes:
  grafana-data:

This configuration allows you to safely back up things like databases, if you can tolerate a bit of downtime.

Pre/post backup exec

If you don't want to stop the container while it's being backed up, and the container comes with a backup utility (this is true for most databases), you can label the container with commands to run before/after backing it up:

version: "3"

services:

  database:
    image: influxdb:1.5.4
    volumes:
      - influxdb-data:/var/lib/influxdb         # This is where InfluxDB keeps its data
      - influxdb-temp:/tmp/influxdb             # This is our temp space for the backup
    labels:
      # These commands will be exec'd (in the same container) before/after the backup starts:
      - docker-volume-backup.exec-pre-backup=influxd backup -portable /tmp/influxdb
      - docker-volume-backup.exec-post-backup=rm -rfv /tmp/influxdb

  backup:
    image: jareware/docker-volume-backup
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro # Allow use of the "pre/post exec" feature
      - influxdb-temp:/backup/influxdb:ro       # Mount the temp space so it gets backed up
      - ./backups:/archive                      # Mount a local folder as the backup archive

volumes:
  influxdb-data:
  influxdb-temp:

The above configuration will perform a docker exec for the database container with influxd backup, right before the backup runs. The resulting DB snapshot is written to a temp volume (influxdb-temp), which is then backed up. Note that the main InfluxDB data volume (influxdb-data) isn't used at all, as it'd be unsafe to read while the DB process is running.

Similarly, after the temp volume has been backed up, it's cleaned up with another docker exec in the database container, this time just invoking rm.

If you need a more complex script for pre/post exec, consider mounting and invoking a shell script instead.

Configuration

Variable	Default	Notes
BACKUP_SOURCES	/backup	Where to read data from. This can be a space-separated list if you need to back up multiple paths, when mounting multiple volumes for example. On the other hand, you can also just mount multiple volumes under /backup to have all of them backed up.
BACKUP_CRON_EXPRESSION	@daily	Standard debian-flavored cron expression for when the backup should run. Use e.g. 0 4 * * * to back up at 4 AM every night. See the man page or crontab.guru for more.
BACKUP_FILENAME	backup-%Y-%m-%dT%H-%M-%S.tar.gz	File name template for the backup file. Is passed through date for formatting. See the man page for more.
BACKUP_ARCHIVE	/archive	When this path is available within the container (i.e. you've mounted a Docker volume there), a finished backup file will get archived there after each run.
PRE_BACKUP_COMMAND		Commands that is executed before the backup is created.
POST_BACKUP_COMMAND		Commands that is executed after the backup has been transferred.
BACKUP_UID	root (0)	After backup file has been moved to archive location the file user ownership is changed to this UID.
BACKUP_GID	$BACKUP_UID	After backup file has been moved to archive location the file group ownership is changed to this GID.
BACKUP_WAIT_SECONDS	0	The backup script will sleep this many seconds between re-starting stopped containers, and proceeding with archiving/uploading the backup. This can be useful if you don't want the load/network spike of a large upload immediately after the load/network spike of container startup.
BACKUP_HOSTNAME	$(hostname)	Name of the host (i.e. Docker container) in which the backup runs. Mostly useful if you want a specific hostname to be associated with backup metrics (see InfluxDB support).
BACKUP_CUSTOM_LABEL		When provided, the start/stop and pre/post exec logic only applies to containers with this custom label.
CHECK_HOST		When provided, the availability of the named host will be checked. The host should be the destination host of the backups. If the host is available, the backup is conducted as normal. Else, the backup is skipped.
AWS_S3_BUCKET_NAME		When provided, the resulting backup file will be uploaded to this S3 bucket after the backup has ran. You may include slashes after the bucket name if you want to upload into a specific path within the bucket, e.g. your-bucket-name/backups/daily.
AWS_GLACIER_VAULT_NAME		When provided, the resulting backup file will be uploaded to this AWS Glacier vault after the backup has ran.
AWS_ACCESS_KEY_ID		Required when using AWS_S3_BUCKET_NAME.
AWS_SECRET_ACCESS_KEY		Required when using AWS_S3_BUCKET_NAME.
AWS_DEFAULT_REGION		Optional when using AWS_S3_BUCKET_NAME. Allows you to override the AWS CLI default region. Usually not needed.
AWS_EXTRA_ARGS		Optional additional args for the AWS CLI. Useful for e.g. providing --endpoint-url <url> for S3-compatible systems, such as DigitalOcean Spaces, MinIO and the like.
SCP_HOST		When provided, the resulting backup file will be uploaded by means of scp to the host stated.
SCP_USER		User name to log into SCP_HOST.
SCP_DIRECTORY		Directory on SCP_HOST where backup file is stored.
PRE_SCP_COMMAND		Commands that is executed on SCP_HOST before the backup is transferred.
POST_SCP_COMMAND		Commands that is executed on SCP_HOST after the backup has been transferred.
GPG_PASSPHRASE		When provided, the backup will be encrypted with gpg using this passphrase.
INFLUXDB_URL		When provided, backup metrics will be sent to an InfluxDB instance at this URL, e.g. https://influxdb.example.com.
INFLUXDB_DB		Required when using INFLUXDB_URL; e.g. my_database.
INFLUXDB_CREDENTIALS		Required when using INFLUXDB_URL; e.g. user:pass.
INFLUXDB_MEASUREMENT	docker_volume_backup	Required when using INFLUXDB_URL.
TZ	UTC	Which timezone should cron use, e.g. America/New_York or Europe/Warsaw. See full list of available time zones.

Metrics

After the backup, the script will collect some metrics from the run. By default, they're just written out as logs. For example:

docker_volume_backup
host=my-demo-host
size_compressed_bytes=219984
containers_total=4
containers_stopped=1
time_wall=61.6939337253571
time_total=1.69393372535706
time_compress=0.171068429946899
time_upload=0.56016993522644

If so configured, they can also be shipped to an InfluxDB instance. This allows you to set up monitoring and/or alerts for them. Here's a sample visualization on Grafana:

Automatic backup rotation

You probably don't want to keep all backups forever. A more common strategy is to hold onto a few recent ones, and remove older ones as they become irrelevant. There's no built-in support for this in docker-volume-backup, but you are able to trigger an external Docker container that includes rotate-backups. In the examples, we draw on docker-rotate-backups.

In order to start an external Docker container, access to docker.sock has to be granted (as already seen in in the section on stopping containers while backing up). Then, docker-rotate-backups can be run on local directories as well as on remote directories.

The default rotation scheme implemented in docker-rotate-backups preserves seven daily, four weekly, twelve monthly, and every yearly backups. For detailed information on customizing the rotation scheme, we refer to the documentation.

Rotation for local backups

Let /home/pi/backups be the path to your local backups. Then, initialize the environmental variable POST_BACKUP_COMMAND with the following command.

environment:
  POST_BACKUP_COMMAND: "docker run --rm -e DRY_RUN=false -v /home/pi/backups:/archive ghcr.io/jan-brinkmann/docker-rotate-backups"
volumes:
  - /var/run/docker.sock:/var/run/docker.sock:ro
  - /home/pi/backups:/archive

Rotation for backups tranferred via SCP

Here, let /home/pi/backups be the backup directory on a remote host. To run docker-rotate-backups on that directory, the command in POST_BACKUP_COMMAND has to include all necessary information in order to access the remote host by means of SSH. Remember, if you transfer your backups by means of SCP, all information in SSH_USER, SSH_HOST, SSH_ARCHIVE, and the SSH public key are already available.

environment:
  SCP_HOST: 192.168.0.42
  SCP_USER: pi
  SCP_DIRECTORY: /path/to/backups
  POST_BACKUP_COMMAND: "docker run --rm -e DRY_RUN=false -e SSH_USER=pi -e SSH_HOST=192.168.0.42 -e SSH_ARCHIVE=/home/pi/backups -v /home/pi/.ssh/id_rsa:/root/.ssh/id_rsa:ro ghcr.io/jan-brinkmann/docker-rotate-backups"
volumes:
  - /var/run/docker.sock:/var/run/docker.sock:ro
  - /home/pi/.ssh/id_rsa:/ssh/id_rsa:ro

Rotation for S3 backups

Amazon S3 has Versioning and Object Lifecycle Management features that can be useful for backups.

First, you can enable versioning for your backup bucket:

Then, you can change your backup filename to a static one, for example:

environment:
  BACKUP_FILENAME: latest.tar.gz

This allows you to retain previous versions of the backup file, but the most recent version is always available with the same filename:

$ aws s3 cp s3://my-backup-bucket/latest.tar.gz .
download: s3://my-backup-bucket/latest.tar.gz to ./latest.tar.gz

To make sure your bucket doesn't continue to grow indefinitely, you can enable some lifecycle rules:

These rules will:

• Move non-latest backups to a cheaper, long-term storage class (Glacier)
• Permanently remove backups after a year
• Still always keep the latest backup available (even after a year has passed)

Testing

A bunch of test cases exist under test. To run them:

cd test/backing-up-locally/
docker-compose stop && docker-compose rm -f && docker-compose build && docker-compose up

Some cases may need secrets available in the environment, e.g. for S3 uploads to work.

Releasing

1. Draft a new release on GitHub
2. docker build -t jareware/docker-volume-backup:x.y.z .
3. docker push jareware/docker-volume-backup:x.y.z