On 16/04/2024 14:20, Fabian Grünbichler wrote:

> if the start is not part of an incoming migration, check the VM against its

> pool's run limit.

> 

> Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>

> ---

>  PVE/QemuServer.pm | 13 +++++++++++++

>  1 file changed, 13 insertions(+)

> 

> diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm

> index be937ec1..78b6ff96 100644

> --- a/PVE/QemuServer.pm

> +++ b/PVE/QemuServer.pm

> @@ -5704,6 +5704,19 @@ sub vm_start_nolock {

>  	    # But we do it above, so for now let's be consistent.

>  	    $conf = PVE::QemuConfig->load_config($vmid); # update/reload

>  	}

> +

> +	my $cpu_count = 1;

> +	$cpu_count = $conf->{sockets} if $conf->{sockets};

> +	$cpu_count *= ($conf->{cores} || 1);

> +	$cpu_count = $conf->{vcpus} if $conf->{vcpus} && $conf->{vcpus} < $cpu_count;

> +

> +	my $changes = {

> +	    absolute => 1,

> +	    running => 1,

> +	    mem => get_current_memory($conf->{memory}),



Bug: The memory value here should be converted from MiB to bytes, else

the pool limit will not be enforced as `check_guest_pool_limit` expects

the value to be in bytes, i.e. `get_current_memory(...) * 1024 * 1024`.



Spotted it by starting a second VM (which would've raised the memory

usage over the limit) and wondered why it didn't fail. For the third VM

I got:



```

pool limits exhausted: (run) mem: 4096 + 0.001953125 over 3072

```



> +	    cpu => $cpu_count,

> +	};

> +	PVE::GuestHelpers::check_guest_pool_limit($vmid, $changes);

>      }

>  

>      # override offline migrated volumes, conf is out of date still

> -- 

> 2.39.2