Tips for shell_command and auto-deleting old snapshots

Tags: #<Tag:0x00007f7398eddb00>

I was having an issue trying to delete old snapshots using shell_command. After a year of being annoyed about it, I finally got it working today. Along the way I figured out a few important things I feel are worth sharing to the community. Fully functional script at the end.

  • shell_command runs in the homeassistant container if you are running HASSIO. I’m not sure about other installation types, but probably consistent. The mounted drives are different in this container vs. the host system that you can access via the developer SSH on port 22222. Keep that in mind when testing via SSH. You’ll want to run a shell inside the homeassistant container to test from or else the paths won’t be right if you’re on the host.

  • The backups are not mounted by default in the homeassistant container, so you have to mount them in your script. In my case, with full HASSIO, I mounted /dev/sda8 on /mnt/data and the snapshots are at /mnt/data/supervisor/backup. You can mount wherever you want. I chose /mnt/data to be consistent with where the host system mounts that device so that I don’t have to remember different locations based on what I’m logged into.

  • Although the backup folder wasn’t mounted for us in the homeassistant container, /config, /share and /ssl are available. The addons folder is not mounted either, but mounting the drive ourselves gives us that too, if you need it for some reason.

  • Note that all of those folders are technically subfolders on /dev/sda8, so when we mounted that drive to /mnt/data they are also accessible through that mount. For example, /config and /mnt/data/supervisor/homeassistant are the same folder on the disk. The other folders are also under the supervisor folder, but their names are consistent (share, ssl, backup, addons). I find it odd that the config folder is called homeassistant on the disk, but whatever.

  • A limitation I ran into with shell_command is that it doesn’t like to run multiple commands, nor does it handle for loops. I think it’s a limitation of the compact shell that runs the command (BusyBox). I worked around this limitation by making a bash script to do all the work and the shell_command just calls my script. I created my bash script in my config folder so that it’s part of my backups. Using a separate bash script file also lets me make changes without having to restart homeassistant, whereas editing the shell_command settings in configuration.yaml would require restarting homeassistant.

  • My bash script automates creating a temporary mount folder, mounting the drive, looping the available *.tar files over 7 days old and deciding which ones to keep or delete, then unmounting and removing the temporary mount folder. Aside from age, the decision to keep or delete requires examining the friendly name of the snapshot to determine if it’s one of the automated backups because I want to retain all manual backups regardless of age. I found a way to get the friendly name from the snapshot.json file inside the tar file without having to write it to the filesystem, so it’s nice and tidy.

My script file /config/tools/


echo "Cleaning up automated snapshots older than 7 days. Retaining manual backups of any age."
mkdir /mnt/data
mount /dev/sda8 /mnt/data
for i in `find /mnt/data/supervisor/backup/*.tar -mtime +7`
  NAME=`tar xfO $i ./snapshot.json | jq .name`
  AGE="$((($(date +%s) - $(date +%s -r $i))/86400))"
  if [[ $NAME == *Automated* ]]; then
    echo "  - removing $(basename $i) from $AGE days ago ($NAME)"
    rm $i
    echo "  + keeping  $(basename $i) from $AGE days ago ($NAME)"
umount /mnt/data
rmdir /mnt/data

To call it, I have this in my configuration.yaml:

  delete_old_backups: '/config/tools/'

My next project is mounting a SAMBA share on my network and sync the snapshots so they are properly backed up on another computer.