out of memory - starting new vm = crashing runnig vm

[Shiba86]

Hello Guys,
hope some know the following problem and maybay there is a solution or a workaround.

We are runnig VE 6.4-4 in our cluster environment.

7 Hosts
Ceph storage

- every single host runnig with 256 GB of memory

We are planning to install 4# MS Exchange Server VMs on this cluster.
Every VM gets 128 GB of memory.

So... now we are comming to the "Bug".

when I start one of the Exchange VMs, all is fine...

when I start a second Exchange VM on the same host, on which the first Exchange VM is runnig, I provoke an "out of memory" situation.
(2* 128 GB memory + host overhead = nor enough memory)

immediately after starting the second Exchange VM, the first VM crashing down without any information or log message.

I was wondering how can this happend on a "professional" solution?!

I expect a message like " you cant start the vm, not enough memory" or I expect a automatic migration task to an host with free memory.


Is this a known bug? Is there a fix or workaround?
Any recomandation?

Thanks for your input!

 

[dcsapak]

this is normal linux kernel behaviour, if you get into an oom sitation, the kernel "solves" it by killing a process
the solution here is not to overcommit your memory

if you are running ceph hyperconverged, to not forget that the osds /monitors need memory themselves, as well as the pve daemons