vzdump Backup Hook Script is ignored

[RockNLol]

hi,
I'm trying to wake my HTPC which hosts a PBS-VM up before starting a backup. So I wrote a little bash script as hook for vzdump:

root@pve:~# cat /usr/bin/backup-hook-script
#!/bin/bash

touch /var/tmp/hooktest # for testing purposes

# INSTALL net rpc COMMAND BY RUNNING:
# apt update && apt install samba-common-bin

# VARIABLES:
WOLUUID='uuid=<opnsense uuid>'
APIKEY="<opnsense apikey>"
SECRET="<opnsense apikey secret>"
HOST="opnsense.mydomain.local"

case "${1}" in

    job-start)
        # wake HTPC that hosts PBS-VM

        # wake via wakeonlan command
        #wakeonlan <macaddress>

        # wake via opnsense API
        curl -XPOST -d $WOLUUID -g -k -u $APIKEY:$SECRET "https://"$HOST":443/api/wol/wol/set/"

        echo "waiting 90s..."
        sleep 90
        echo "PBS should be up now. Continuing..."
    ;;


    job-end)
        net rpc -S <HTPC-IP> -U <user>%<supersecretpassword> shutdown -t 1 # shut HTPC down again
    ;;
esac

... and modified /etc/vzdump.conf:

root@pve:~# cat /etc/vzdump.conf
# vzdump default settings

#tmpdir: DIR
#dumpdir: DIR
#storage: STORAGE_ID
#mode: snapshot|suspend|stop
#bwlimit: KBPS
#ionice: PRI
#lockwait: MINUTES
#stopwait: MINUTES
#size: MB
#stdexcludes: BOOLEAN
#mailto: ADDRESSLIST
#maxfiles: N
script: /usr/bin/backup-hook-script
#exclude-path: PATHLIST
#pigz: N

if i run /usr/bin/backup-hook-script job-start or /usr/bin/backup-hook-script job-end the HTPC/PBS wakes up and shuts down correctly, so my script should be working. To see if the script even gets executed I added the line touch /var/tmp/hooktest # for testing purposes, but theres no /var/tmp/hooktest file created when I run the backup. I get TASK ERROR: could not activate storage 'PBS': PBS: error fetching datastores - 500 Can't connect to pbs.mydomain.local:8007 (No route to host) straight away.

What did I do wrong?

 

[fabian]

activating the storage happens before the hook script is called..

 

[RockNLol]

Well, thats a fundamental issue in my plan then. I guess I have to wake it up by schedule then, or does anybody have a better idea?

 

[fabian]

if you just want this for scheduled backups, you can also adapt the generated cron entry to call your script before/after the vzdump call, but you lose the ability to edit that job via the API/GUI then..

 

[RockNLol]

Thats a good idea. Not being able to edit it via GUI/API is not a problem, scheduled backup should be a "setup once and leave it" kind of deal anyways.

*another idea: wake the PBS up by schedule and trigger the backup on the PVE remotely from a script on the PBS. I'll have a look into the API in the evening.

 

[Jarek Hartman]

activating the storage happens before the hook script is called..

To me there should be another hook calling out a script before storage activation is invoked.

Pretty frustrating that the flow with Wake-on-lan is not possible off the shelf

 

[fabian]

there is nowadays (well, in the next pve-manager version bump ) - the phase is called `job-init`, and the storage config is reloaded after it has been called to pick up any potential changes by the hookscript. so you can enable a storage, do WOL, wait for it to come up, and only then return from the hook script for the backup to proceed.

 

[adrians]

Thank you, Fabian. I have a parallel requirement that this should meet.
At the moment I have a bunch of "utility" scripts, which I run inside a docker container (for historic reasons) inside a VM on pve.
A crontab runs a python script that does a bunch of sshs into the pve host to do a wake-on-lan, mount the storage, un-disable it, run the backup, disable the storage and unmount it.

The disabling and unmounting is necessary to avoid a mountpoint persisting on the host for a machine that is asleep, which can result in bad things happening.

One thing that I had to solve was to generate the list of vms/cts to be backed-up. I was looking for a way to interrogate a disabled backup job, but didn't find it. The "hack" solution I found was to create a "Backup_Role" and backup group, and assign permission to "Backup_Role" in vm/ct. This can be read via pvesh get /access/acl on the host.

It will be much cleaner if the hook scripts can remove my need for this workaround and hack.

 

[fabian]

the job itself is not modifiable by the hookscript (at least not in a way that would be picked up by the current execution). if you need this kind of flexibility, not using the built-in scheduler and instead writing your own wrapper around vzdump (the CLI, or the API endpoint) and scheduling that via a systemd timer or cron job seems like a good solution.