[pve-devel] [Veeam] Veeam change requests?

[Dominik Csapak]

On 9/17/24 09:20, Andreas Neufert via pve-devel wrote:
> 
> Hi Proxmox Dev team,
> 

Hi,

> Tim Marx mentioned that you have some insights and change wishes for the Veeam backup processing and that we should reach out to this list. We would be happy to get this feedback here to be able to address it in our code or join a call if this helps.

Thanks for reaching out!

During (very basic & short) testing, i discovered a few things that are problematic from our point 
of view:

* During backup, there is often a longer running connection open to our QMP socket of running VMs
   (/var/run/qemu-server/XXXX.qmp, where XXXX is the vmid). This blocks our management stack from
   doing certain tasks, like start/stop (probably does not matter during backup) but also
   things like the VNC console, etc.

   a better way would be to close the connections as soon as possible instead of keeping them
   open. (Alternatively using our API/CLI could also be done, but I don't know what
   exact QMP commands you're running)

   if you absolutely need a longer running socket, please open a bug report on
   https://bugzilla.proxmox.com so we can discuss and track that there, how we could make
   a socket available that is not used by our stack

* Another thing that I noticed was that it's not really visible if a backup is running
   for a particular VM, so users might accidentally them down (or pause, etc.). Especially
   i think it's bad if the VM is placed under a HA policy that has 'stopped' as target, as
   that will try to stop the VM by itself. (Though this might be a configuration error in itself?)

   A quick way to fix this would be to have a (custom) lock in our VMs. For longer running tasks
   that block a guest, we have a line 'lock: XXXX' in the config that prevents our stack
   from doing most operations.

   Putting that in would be a very short call to our perl code that locks the config locally
   ( `PVE::QemuConfig->lock_config($vmid, $updatefn) ), checks for existing locks,
   updates the config with a new (custom) lock and writes it again.

   Though i must admit, I'm not sure if custom locks outside of our defined ones would work,
   but I'm sure we could add a 'custom' lock that you could use, should my mentioned
   approach not work properly.

* Also, I noticed that when a guest is started from your stack, you modify the QEMU command line a
   bit, namely removing some options that would be necessary to start the VM during the backup.
   Is there a specific reason why you do it this way, instead of starting the VM through
   our API/CLI?


A more general question last: What is the process for our/your users and us if they/we
find a bug? Where can they be reported to you?

I hope this helps

Best regards
Dominik