On Mon, 2021-03-08 at 10:52 +0000, Daniel P. Berrangé wrote:
On Fri, Mar 05, 2021 at 08:14:02PM +0100, Andrea Bolognani wrote:
> +# If enabled, libvirt will not attempt to change process limits (as
> +# configured with the max_processes, max_files and max_core settings
> +# below) itself but will instead expect an external entity to perform
> +# this task.
Can't users simply not set max_core, max_files, etc already ?
That works for things that are static and have a corresponding
configuration option in qemu.conf, but the memory locking limit is
dynamic, per-VM and needs to change as devices are added and removed
from the guest.
I think it is preferrable to have flags tailored specifically to
the individual limits, not a global flag. Otherwise you can end
up in a case where you want to disable the memory limits, but
keep the other limits set which is impossible with this global
flag.
Since what I'm interested in is the memory locking limit, I guess I
could turn this into
max_memlock_external = 1
or even
max_memlock = "external"
with "dynamic" being the other accepted value, which would be the
default and would behave as libvirt does today.
Do you think that would work better?
--
Andrea Bolognani / Red Hat / Virtualization