VM not auto starting and also strange behaviour

[albgen]

Hello,

i have 4 VM on my proxmox host. Tried to reboot the whole host but only 3 of them started.
The one that did not started has actually "Start at boot = yes", Start Order 2, Start Delay 60...

The strange behaviour is that when i start manually the machine, the other get automatically shutdown.
from the logs i see that it is killed because it is running out of memory but i think there is plenty of memory on the server.

thank you


Logs

https://pastebin.pl/view/546e9f71

 

[fiona]

Hi,
how much memory do you have on the host and how much memory is assigned to the VMs? You should also check how much memory is actually used on the host using free, top,etc., or simply check the total usage in the Summary tab for your node in the GUI.

 

[albgen]

hi,

proxmox was right, i mean the RAM was less than the total needed to run the VMs. The issue though is that, proxmox should not attempt to start a VM when there is no ram available.

The current behavior, which means it will kill a random vm when there is no ram, is very dangerous. I mean i tried to start even manually, and it killed another vm.
In environment with a lot of VMs, you cannot keep the maths of how much ram is free and how is occupied.

Does it make sense?

thank you

 

[fiona]

hi,

proxmox was right, i mean the RAM was less than the total needed to run the VMs. The issue though is that, proxmox should not attempt to start a VM when there is no ram available.

With things like ballooning, it is often possible to run VMs where the memory provisioned to VMs sums up to more than available on the host. One would need to have a way to see in advance to see how much the kvm process is actually going to use before starting the VM.

The current behavior, which means it will kill a random vm when there is no ram, is very dangerous. I mean i tried to start even manually, and it killed another vm.
In environment with a lot of VMs, you cannot keep the maths of how much ram is free and how is occupied.

That's why there is tools to monitor the RAM usage (also in the node's Summary page).

Does it make sense?

thank you

 

[albgen]

With things like ballooning, it is often possible to run VMs where the memory provisioned to VMs sums up to more than available on the host. One would need to have a way to see in advance to see how much the kvm process is actually going to use before starting the VM.

ok, but dont kill a random machine...just be more intelligent..kill the one that at least is in starting status...

 

[fiona]

It's not random, but consuming lots of memory makes it more likely for a process to get killed. Maybe the newly started VM didn't consume much memory already, so it wasn't chosen? Here are more details about how the OOM killer in Linux works.