Backup a docker-compose project, including all images, named and unnamed volumes, container filesystems, config, logs, and databases.

docker-compose-backup.sh

#!/usr/bin/env bash

### Bash Environment Setup
# http://redsymbol.net/articles/unofficial-bash-strict-mode/
# https://www.gnu.org/software/bash/manual/html_node/The-Set-Builtin.html
# set -o xtrace
set -o errexit
set -o errtrace
set -o nounset
set -o pipefail
IFS=$'\n'

# Fully backup a docker-compose project, including all images, named and unnamed volumes, container filesystems, config, logs, and databases. 
project_dir="${1:-$PWD}"
if [ -f "$project_dir/docker-compose.yml" ]; then
    echo "[i] Found docker-compose config at $project_dir/docker-compose.yml"
else
    echo "[X] Could not find a docker-compose.yml file in $project_dir"
    exit 1
fi

project_name=$(basename "$project_dir")
backup_time=$(date +"%Y-%m-%d_%H-%M")
backup_dir="$project_dir/data/backups/$backup_time"

# Source any needed environment variables
[ -f "$project_dir/docker-compose.env" ] && source "$project_dir/docker-compose.env"
[ -f "$project_dir/.env" ] && source "$project_dir/.env"


echo "[+] Backing up $project_name project to $backup_dir"
mkdir -p "$backup_dir"

echo "    - Saving docker-compose.yml config"
cp "$project_dir/docker-compose.yml" "$backup_dir/docker-compose.yml"

# Optional: pause the containers before backing up to ensure consistency
# docker compose pause

# Optional: run a command inside the contianer to dump your application's state/database to a stable file
echo "    - Saving application state to ./dumps"
mkdir -p "$backup_dir/dumps"
# your database/stateful service export commands to run inside docker go here, e.g.
#   docker compose exec postgres env PGPASSWORD="$POSTGRES_PASSWORD" pg_dump -U "$POSTGRES_USER" "$POSTGRES_DB" | gzip -9 > "$backup_dir/dumps/$POSTGRES_DB.sql.gz"
#   docker compose exec redis redis-cli SAVE
#   docker compose exec redis cat /data/dump.rdb | gzip -9 > "$backup_dir/dumps/redis.rdb.gz"

for service_name in $(docker compose config --services); do
    image_id=$(docker compose images -q "$service_name")
    image_name=$(docker image inspect --format '{{json .RepoTags}}' "$image_id" | jq -r '.[0]')
    container_id=$(docker compose ps -q "$service_name")

    service_dir="$backup_dir/$service_name"
    echo "[*] Backing up ${project_name}__${service_name} to ./$service_name..."
    mkdir -p "$service_dir"
    
    # save image
    echo "    - Saving $image_name image to ./$service_name/image.tar"
    docker save --output "$service_dir/image.tar" "$image_id"
    
    if [[ -z "$container_id" ]]; then
        echo "    - Warning: $service_name has no container yet."
        echo "         (has it been started at least once?)"
        continue
    fi

    # save config
    echo "    - Saving container config to ./$service_name/config.json"
    docker inspect "$container_id" > "$service_dir/config.json"

    # save logs
    echo "    - Saving stdout/stderr logs to ./$service_name/docker.{out,err}"
    docker logs "$container_id" > "$service_dir/docker.out" 2> "$service_dir/docker.err"

    # save data volumes
    mkdir -p "$service_dir/volumes"
    for source in $(docker inspect -f '{{range .Mounts}}{{println .Source}}{{end}}' "$container_id"); do
        volume_dir="$service_dir/volumes$source"
        echo "    - Saving $source volume to ./$service_name/volumes$source"
        mkdir -p $(dirname "$volume_dir")
        cp -a -r "$source" "$volume_dir"
    done

    # save container filesystem
    echo "    - Saving container filesystem to ./$service_name/container.tar"
    docker export --output "$service_dir/container.tar" "$container_id"

    # save entire container root dir
    echo "    - Saving container root to $service_dir/root"
    cp -a -r "/var/lib/docker/containers/$container_id" "$service_dir/root"
done

echo "[*] Compressing backup folder to $backup_dir.tar.gz"
tar -zcf "$backup_dir.tar.gz" --totals "$backup_dir" && rm -Rf "$backup_dir"

echo "[√] Finished Backing up $project_name to $backup_dir.tar.gz."

# Resume the containers if paused above
# docker compose unpause

I am very fresh to docker and compose as well, but I managed to offload my Home Assistant with some add ons using another linux machine with docker compose
currently I am running there:

docker compose config --services
chrony
duckdns
grafana
mariadb
mosquitto
node-red
phpmyadmin
pihole
timemachine
wireguard

now, I am trying to find some good backup solution. and this is how I landed here.
I took your script, modified where needed "docker-compose" to "docker compose" and voila - all works.
But I have 3 questions:
1- my "timemachine" container uses "/mnt/timemachine" volume where the data is stored - apparently we are talking GB here, so I would like to exclude this container from the backup script - is there any way here to exclude it?
2- when database is backed up apparently it would be nice to stop it, right? not necessarily other containers - is there any easy way to do it in this script?
3- in case I stop the container (docker compose stop mariadb) - would simple command: "tar cfz maria.tar.gz /srv/docker/mariadb" be enough to really have EVERYTHING backed up for this container (of course, also the docker-compose.yml)?
what are all these "dumps", "logs" etc - isn't everything in the volume itself? provided my only volumes are inside the same folder, in this example /srv/docker/mariadb

thank you for your help ;)

1. to exclude a container you'd modify this line to add a filter to the list of containers it loops through, e.g.

- for service_name in $(docker compose config --services); do
+ for service_name in $(docker compose config --services | grep -v timemachine); do

2. this is up to you, pgdump and redis SAVE don't require you pause the container to ensure consistency, but if you need to pause your other containers to ensure consistency at the application layer, I did mention that in these lines already:

# Optional: pause the containers before backing up to ensure consistency
# docker compose pause

3. no, that only backs up some of that containers state, there is other hidden state in the docker system like remote mounts, stdout/stderr log output, container config and environment state, etc. which is why this script exists in the first place. you can read the state it generates on :67, :71, etc. to see how it's separate from the volume contents

@pirate i need your help, i customized the script and executed the one you shared as is, and my docker is gone, how do i reverse this

Awsome script, could you edit it so you can add environment variables for containers in the docker compose file so you can specify if you want to make backups of the attached volumes or not? For Example when backing-up Plex i dont want to backup the volume that contains my music and movies.

I add this feature : https://gist.github.com/GeoHolz/ee9362c82ee13f8a5690d86d6ec7bb0c
Thanks pirate !

bak.sh: 8: set: Illegal option -o errtrace
I assume you invoke the script with sh bak.sh. Be sure to use bash bak.sh

I have backed up other docker containers, but found Bookwyrm exposed some issues with others may also encounter - I am not fully there yet, so I will post in individual comments for each issue.

First a neat snippet from the Bookwyrm bw-dev script

if docker compose &> /dev/null ; then DOCKER_COMPOSE="docker compose" else DOCKER_COMPOSE="docker-compose" fi
and then use DOCKER_COMPOSE in the rest of the script, unifying for Debian based and other systems.

My .env had a variable (a passkey) which contained a $ (dollar) sign, which caused sourcing it to fail the shell could not find the variable which was the random string following the $

Worked around it for now, simply by commenting out that line in .env - to move on past it, but not an ideal situation - I know (as of very recently - not really a docker person) that secrets would be a better way, but this would be a big upstream change,

Bind volumes - attempting to back up Bind Volumes caused the script to fail -
adapted the save data volumes part to

# save data volumes
mkdir -p "$service_dir/volumes"
for typesource in $(docker inspect -f '{{range .Mounts}}{{println .Type .Source}}{{end}}' "$container_id"); do
type=$(echo $typesource | awk '{print $1}')
if [[ $type == "volume" ]]; then 
  source=$(echo $typesource | awk '{print $2}')
      volume_dir="$service_dir/volumes$source"
      echo "    - Saving $source volume to ./$service_name/volumes$source"
      mkdir -p $(dirname "$volume_dir")
      cp -a -r "$source" "$volume_dir"
 else
	  echo "    - not backing up volume $source with type $type"
fi
 done

One of the services did not have an image - protect against this case with

for service_name in $(docker-compose config --services); do
image_id=$(docker-compose images -q "$service_name")
if [[ -z "$image_id" ]]; then
echo " - Note: $service_name has no image"
else
image_name=$(docker image inspect --format '{{json .RepoTags}}' "$image_id" | jq -r '.[0]')
fi
container_id=$(docker-compose ps -q "$service_name")

and
# save image
if [[ ! -z "$image_id" ]]; then
echo " - Saving $image_name image to ./$service_name/image.tar"
docker save --output "$service_dir/image.tar" "$image_id"
fi

This enabled my docker-compose backup to complete - I will now copy it its migration target host and try a restore on there. Although I
take your point at https://gist.github.com/pirate/265e19a8a768a48cf12834ec87fb0eed#gistcomment-3525603, from experience I know that while backups are nice to have, what people really want is restores - and for that a backup is a required starting point.