Copy backup from PBS datastore to S3FS

[mlazzarotto]

I'm using an S3-compatible cloud storage (Idrive E2) and I'd like to use it to keep a copy of my backups.
I've mounted the bucket using S3FS, but since I can't use that as a PBS datastore, I was wondering if I can extract the backups from PBS and copy them to the S3FS mounted folder in a similar way as I would copy a dump from PVE.

I do have the option to sync all the PBS files (chunks) but my datastore (1.5TB) is bigger than my account capacity (1TB), so that's not an option.

 

[ggoller]

Hi!
this is not really possible, as the backups are stored in chunks, which are deduplicated. This means that multiple backups can refer to the same chunk; so you have to copy the whole chunk store, otherwise it won't work. (https://pbs.proxmox.com/docs/technical-overview.html.)

But I would add the s3fs mount as a datastore to PBS, then sync some backups from one datastore to the s3fs one.

 

[mlazzarotto]

But I would add the s3fs mount as a datastore to PBS, then sync some backups from one datastore to the s3fs one.

I wish I could, but there's something wrong with the Idrive implementation of their S3 "version" that messes up with the creation of the chunks.

Anyway, I guess my next step will be to add the s3fs mount to my Proxmox VE and use that to store the backups (in addition to PBS).

 

[mlazzarotto]

@ggoller
The s3fs mount may fail for the most weird issue. To prevent the backup job from filling the disk space, I think that I can use a hook script to check the mount. If the mount fails, how can the hook script terminate the backup job?

 

[ggoller]

This should be possible by just `die`'ing in perl. See https://perldoc.perl.org/functions/die.

 

[khe]

Hi,

most bucket based object storages like S3 or B2 do not support the concept of empty directories (according to rclone mount). Thus if you try to create a chunk, but the folder is not there so it fails.
If I create the missing folders, my sync jobs work, otherwise I get an error, of not being able to insert a chunk.
Not sure, if the hook script is the best way or just run mkdir {{0..9},{a..f}}{{0..9},{a..f}}{{0..9},{a..f}}{{0..9},{a..f}} in the .chunks folder via cron job.

Best regards,
Karl-Heinz

 

[mlazzarotto]

This should be possible by just `die`'ing in perl. See https://perldoc.perl.org/functions/die.

Thanks

 

[mlazzarotto]

Hi,

most bucket based object storages like S3 or B2 do not support the concept of empty directories (according to rclone mount). Thus if you try to create a chunk, but the folder is not there so it fails.
If I create the missing folders, my sync jobs work, otherwise I get an error, of not being able to insert a chunk.
Not sure, if the hook script is the best way or just run mkdir {{0..9},{a..f}}{{0..9},{a..f}}{{0..9},{a..f}}{{0..9},{a..f}} in the .chunks folder via cron job.

Best regards,
Karl-Heinz

Thanks, I'll try to create the folders

 

[mlazzarotto]

I ended up using a hook script made in shell to check if the folder is mounted and stop the backup job if is not.

#!/bin/bash

mountpoint="/mnt/idrive_e2"
trimmed_dumpdir=$(dirname "$DUMPDIR")

# if backup is starting and if mountpoint is not mounted and if we're backing up to Idrive
if [[ $1 == "backup-start" ]] && ! mountpoint -q "$mountpoint" && [[ "$mountpoint" == "$trimmed_dumpdir" ]]; then
    echo "Error: $mountpoint is not a mounted mountpoint. Killing vzdump"
    # kill vzdump process
    killall -q -r "task UPID:\w+:[a-zA-Z0-9]+:[a-zA-Z0-9]+:[a-zA-Z0-9]+:vzdump:\d+:\w+@\w+"
fi

 

[VictorSTS]

Maybe I misunderstood the problem, but there's the is_mountpoint property of a directory storage [1]. When enabled, if the directory isn't mounted the storage will be disabled automatically and a backup to that storage will not be allowed to start.

[1] https://pve.proxmox.com/pve-docs/pvesm.1.html