proxmox "farm" instead of cluster ?

[RolandK]

is it possible to run multiple promox nodes centrally managed more likely to build up a "farm" then a cluster?

for example running one farm master / primary node which is always up and running and all other nodes are running on a temporary basis and only getting started on demand?

i don't need HA and there is no shared storage, all i want is central management and shuffling around virtual machines between nodes

i know i can set ""pvecm expected 1" but i'm not entirely sure what can go wrong when doing that

 

[leesteken]

is it possible to run multiple promox nodes centrally managed more likely to build up a "farm" then a cluster?

As I understand it: the Proxmox staff is working on it but it's not yet in the GUI.

for example running one farm master / primary node which is always up and running and all other nodes are running on a temporary basis and only getting started on demand?

I don't know if that's part of it.

i don't need HA and there is no shared storage, all i want is central management and shuffling around virtual machines between nodes

I do believe that's what they are working on and the command line/API already supports such things but I'm can't be entirely sure of course.

i know i can set ""pvecm expected 1" but i'm not entirely sure what can go wrong when doing that

Split-brain and probably other issues...

 

[pille99]

why should anybody in a professional environment want such configuration ?

 

[ness1602]

Well if you have 10 clients with one node, maybe you want to managed them together,or something like thath

 

[RolandK]

> why should anybody in a professional environment want such configuration ?

why should anybody get a high availability cluster if he doesn't need one?

 

[RolandK]

> Split-brain and probably other issues...

what are the other possible issues ?

what about increasing the number of votes of the first , ever running node ? ( corosync-quorumtool ...)

wouldn't "Last-man-standing" corosync algorithm better fit here? ( https://manpages.debian.org/unstable/corosync/votequorum.5.en.html )

 

[RolandK]

i have set

quorum {
  last_man_standing: 1
  last_man_standing_window: 10000
  provider: corosync_votequorum
}

and it seems to work


use at your own risk and do NOT use that in production!!!

 

[jalict]

> why should anybody in a professional environment want such configuration ?

why should anybody get a high availability cluster if he doesn't need one?

Build and Render farms comes to mind.

We have a system for orchestrating our Proxmox cluster and it is fully managed by a Python service that spins up on-demand ephmeral Build VMs across our nodes and spreads out resource usage across the cluster.

I do not care about HA in this context.

 

[RolandK]

are there other solutions for running multiple servers as a "farm" ?

what about the following idea ?

farm "master" is a stand alone proxmox host/installation on some tiny/energy saving system like intel nuc or whatever. which has nothing but empty PVE vm's running. you may be able to downsize them by stopping services not needed....

you start with PVE VM1 and setup initial clustering with that.

then you add one physical host together with another virtual PVE. so you have three nodes.
same with the next physical host - that makes 5: 3 virtual, 2 physical

with this, you always can shutdown physical nodes but leave the VMs up and running - and you always have quorum

what can go wrong here?

ok, the vm host can crash or have a defect and then all is down - but we speak about FARM and not HA. just for testing or distributing load to multiple machines....

 

[LnxBil]

Generally, If you only want one management GUI and no HA-Failover or HA-Storage, you can just do that and set expected votes to 1.

what can go wrong here?

What if the node with the virtual machines fail? e.g.

HW A: VM 1, VM 2
HW B: VM 3

if HW A fails, you will not be able to start the failed VM on HW B.

 

[RolandK]

>What if the node with the virtual machines fail

then all fails

i want centralized management and want to be able to use dynamic number of phyiscal hosts

i can live without high availability in testing environment

 

[LnxBil]

then all fails

Yes, that would be unaccpetable for me, yet if you could work with it, then do it.