Paperless NGX backup / Copying is forcing me to have a bigger disk than needed

[ProfDrYoMan]

Hi,

do not want to open an issue since this is no real bug.

See here:

ProxmoxVE/ct/paperless-ngx.sh

Line 43 in 9710f32

cp -r "/opt/paperless/$dir_name" "$BACKUP_DIR/" 2>/dev/null || true

This copies the whole data (which is a lot for my setup) and forces me to double the space for the LXC disk.

Why not moving data aside and then back after update?

Best regards.

PS: @MickLesk, tagging you, since you did change a lot in there the last days. Thank you for taking care to make this even better.

[MickLesk]

mv is dangerous in the update process because it moves the original data and can lead to partial or complete data loss in the event of any error (abortion, OOM, full disk, defective release). cp -r, on the other hand, creates a complete copy and leaves the original data untouched, allowing for a safe rollback at any time. If /opt and /tmp are located on different file systems, mv internally degrades to “copy + delete” anyway and becomes even more risky. An aborted mv leaves behind a broken data directory – an aborted cp does not. The updater deletes /opt/paperless during deployment; with mv, the original data would already be gone.

[ProfDrYoMan]

And a note on the possibility that the script gets stopped or aborted. In that case the use of $$ in the temp dir will not make the rollback save in all cases, right? The PID will be different for a second run. At least there will be an orphaned copy of the backup on disk.

[ProfDrYoMan]

Oh. I only now see that BACKUPDIR is on /tmp.

This is even more dangerous for me (and possible others) since /tmp is tmpfs and is in RAM, or am I wrong?

root@pl:/# df -h
Filesystem                    Size  Used Avail Use% Mounted on
...
tmpfs                          16G  4.0K   16G   1% /tmp

I do not think creating the backup there is a good idea. My data size is in that area already and still growing.

Please at least stick to making the copy in /opt.

[MickLesk]

Depending on the LXC setup, /tmp is either a real tmpfs or a normal directory on the root FS.

In most Debian 12/13 containers, /tmp is located on / as usual – as is the case with me, which is why the backup does not consume any RAM, even with large paperless data.

In your container, however, /tmp is mounted as tmpfs, so the copy actually ends up in RAM.

If you want to change this, you are welcome to submit a PR that moves the backup path to /opt or /var/lib, for example. For standard containers, however, the current path works reliably.

[ProfDrYoMan]

Thanks for the feedback.

The container was build by the helper-script. I did not manually set /tmp to tmpfs. So I would not expect that it's normal that /tmp goes to a normal disk.

I asked a friend who is using pl-ngx just recently installed with the script. His /tmp is also in tmpfs.

I can do a PR on that, but since this would go thru your hands anyhow, do you mind just changing to create the backup in /opt/paperless_backup_$$ again?

Sticking to copy is understandable, but using RAM is not. :)

[ProfDrYoMan]

Did a PR here: #9195