Configure or disallow concurrent backups

[Carlos Gomes]

Hello everyone

I have a virtual PBS inside one of my servers, on a 4 clustered array.
I noticed if I select a big list of VMs to start a backup on the weekends, since they are on different servers they will start up simultaneously, even though the backup schedule is done on the datacenter global configs.

I know that I could create 4 separated backup schedules, with proper vms for each server, but I would like to ask first if is there a way better to setup backups that would fire up the scheduled list, but normally doing one VM each time.

 

[datschlatscher]

There is no option right now to "automatically coordinate" backup tasks on different nodes in a cluster. If the backup schedules start at the same time, they will all try to backup concurrently. In your case it would probably be best to set up the backup jobs delayed, e.g. by 8 hours per node.

It kind of sounds like you are selecting the VMs for each node manually, if that is the case, you can select the cluster node and then select all guests with a single click. That should make setup of the backup jobs faster.

 

[spirit]

https://bugzilla.proxmox.com/show_bug.cgi?id=3086

 

[Carlos Gomes]

https://bugzilla.proxmox.com/show_bug.cgi?id=3086

hmm thanks spirit, I really thought that this as a think that could be improved, instead of creating schedules manually for each node, since takes some trial and following to proper setup backup times to not overlap, considering the classic "backup everything on the weekend" scheduling.

For now its possible to add a backup chosing all nodes, but if its on a shared store between all of then the queue to not fire up lots of concurrent writes to PBS would be a good solution to guarantee backup integrity and not fry up the PBS storage

Considering using only one PBS shared between the nodes, when its going up scale problems will surely arise.

 

[eXtremeSHOk]

has any progress been made towards this ? It is pretty easy to saturate a 20Gbit link with backups from a cluster and the only solution is to limit the number of concurrent nodes.