I am trying to setup a restic job to backup my docker stacks, and with half of everything owned by root it becomes problematic. I’ve been wanting to look at podman so everything isn’t owned by root, but for now I want to backup my work I built.
Also, how do you deal with some docker containers having databases. Do you have to create exports for all docker containers that have some form of database?
I’ve spent the last few days moving all my docker containers to a dedicated machine. I was using a mix of NFS and local storage before, but now I am doing everything on local NVME. My original plan was having everything on NFS so I would worry about backups there, and I might go back to that.
Don’t backup the container!!
Map volumes with your data to physical storage and then simply backup those folders with the rest of your data. Docker containers are already either backed up in your development directory (if you wrote them) or GitHub so like the operating system itself, no need tk backup anything. The whole idea of Docker is the containers are ephemeral. They are reset at every reboot.
This is the only correct answer.
Containers are ephemeral and stateless. If you’re not mounting a volume, think of what happens if the container dies during your process for exporting the data. A failure mode not possible if you mount.
I have a backup-script (running as root) that
- stops all containers
- runs the restic-backup
- starts all containers
A simpler method would be to stop/start the Docker daemon instead of containers, it works smoothly.
Have a look at https://github.com/minituff/nautical-backup, it does a similar thing
I had this before but this created struggles with some containers since they do start specific checks and scans during startup which resulted in high cpu and disk load.
Since unraid supports zfs, i am using this for the docker stuff and do snapshots to external disk as backup
no need to stop containers anymorw
If you work with databases it’s still safer to stop incoming data for the time of the backup. I don’t know why a higher CPU load would be a problem, those checks don’t run long or do so much your system would be under much stress. Do your backups at 3am if you still think the minute of highe load would cause any problems.
I build my own docker images. All my images are build to run as set id / guid when specified in ansible.
This way only my servicedaemon can do stuff. Also makes sure i never have issues with borgbackup etc.
Don’t run docker as root, don’t run containers as root, pretty simple to be honest.
Can’t you run a restic container where you mount everything? If the restic container is insecure, everything is of course.
But yes, I also migrated to rootless Podman for this reason and a bunch of others.
KISS method: Script that copy the data on the fly to the /tmp dir, compress it, encrypt it and move it to destination using rclone. Running every hour, 4 hours or 24 hours, depending the container.
Never fails. The backups nor the restore.
I mean, snapshotting and piping it to an rclone mount is arguably simpler than trying to do your own ad hoc file syncronization, also does not require 2x the storage space.
For backups I use Nautical Backup.
For the “owned by root” problem, I ensure all my docker compose files have [P]UID and [P]GID set to 1000 (the user my docker runs under). All my 20 containers have no issue running like this.
How are you launching your containers? Docker compose is the way, I have set the following in all mine:
environment: - PUID=1000 - PGID=1000 user: 1000:1000Hey, this is where I am stuck just now: I want to keep the docker volumes, as bind mounts, also on my NAS share. If the containers run as a separate non root user (say 1001) then I can mount that share as 1001… sounds good right?
But somebody suggested running each container from their own user. But then I would need lots of differently owned directories. I wonder if I could keep mounting subdirs of the same NAS share, as different users, so each of them can have their own file access? Perhaps that is overkill.
(For OP: I’ve been on a selfhosting binge the past week and trying to work my way in at least the general direction of best practice… At least for the container databases I’ve been starting to use tiredofit/docker-db-backup (does database dumps) but also discovered this jdfranel docker backup as well which looks great as well. I save the dumps on a volume mounted from NAS. btrfs and there is a folder replication (snapshots) tool. So far, so good. )
…running each container from their own user…
Ideally this is the perfect option from a security standpoint, this as well as each container having it’s own network too.
In a homelab it’s not really required unless you are exposing your network to the internet or are better at creating/managing containers.
If you are just starting out, just keep everything simple.
Do u add bothe the user and env variables or just one?
I add both because why not. It doesn’t hurt.
Do u add bothe the user and env variables or just one?
I used to let root start the backup job
I have my config and data volumes mounted to a share on truenas, that share replicates its snapshots to another truenas server. This is likely not ideal for everyone, but it works for me. My friend that also uses docker has it backed up with duplicati.
First, I try not to have it owned by root. But some containers have special privileges that need to be followed.
So rsync -O will copy the directory retaining permissions and ownership of all files.
I converted they to use Podman.
From my basic selfhosted experience… I run kopia as root , my shit uses bind mounts so all I care about is in that directory.
And so far it works fine, to just down old, rename the directory, copy from nightly backup back the directory and start container.
But yeah if there is something I care about I schedule database dumps like here in bookstack or vaultwarden…
To have something more if shit would not work start.
I do this at the file system level, not the file level, using zfs.
Unless the container has a database, I use zfs snapshots. If it has a database, my script dumps the database first and then does a ZFS snapshot. Then that snapshot is sent via sanoid to a zfs disk that is in a different backup pool.
This is a block level backup, so it only backs up the actual data blocks that changed.
only backup of the data i need to backup (mapped volumes).
Restore: create fresh container and map volumes again.
ZFS snapshots combined with replication to another box. That and a cronjob on packaging up my compose/config files.