Re: [Xen-devel] [PATCH 2/2] expert_mode: Add a new configuration option for expert users.

[George Dunlap <George.Dunlap@xxxxxxxxxxxxx>]

On Wed, Jul 17, 2013 at 6:17 PM, Konrad Rzeszutek Wilk
<konrad.wilk@xxxxxxxxxx> wrote:
> On Wed, Jul 17, 2013 at 11:56:24AM +0100, Ian Jackson wrote:
>> Ian Campbell writes ("Re: [Xen-devel] [PATCH 2/2] expert_mode: Add a new 
>> configuration option for expert users."):
>> > On Fri, 2013-07-12 at 14:08 -0400, Konrad Rzeszutek Wilk wrote:
>> > > This could also be called 'seatbelt' option.
>> > >
>> > > libxl has a variety of checks where it will fail out an operation
>> > > unless the user has provided an --force (or --ignore) parameter.
>> > > Currently one such check is for the 'vcpu-set' command which
>> > > will error out if the count of virtual cpus is greater than the
>> > > physical cpus. This parameter will ignore such checks and allow
>> > > the user to do the operations without the need for override flags.
>> >
>> > Does this overlap  somewhat with various commands which individually
>> > take a -f(orce) option?
>>
>> Clearly it should disable all of those -f's too.
>>
>> > > +=item B<expert_mode=BOOLEAN>
>> > > +
>> > > +Do not act on host performed checks that might lead to performance
>> > > +degradations. Currently checks are made for following operations:
>> > > + - C<vcpu-set> - if the number of VCPUs set for a guest is higher than 
>> > > the
>> > > +   physical count the operation will error out.
>>
>> I don't think this is a very coherent specification.  Surely it should
>> override "all -f options" or something similar.
>
> Sure, I hadn't looked at the other ones.
>>
>> I still don't see why you would want such a thing.
>
> We discussed it during Xen 4.3 release that certain changes, like this one
> are inconsistent (for example you can launch an guest with more vCPUs
> than pCPU, but you if you lower the amount of vCPUs you can't increase
> it until you use -f flag).
>
> But you could consider such inconsistent behavior to be a failsafe
> mechanism so that the user does not do something silly. But if they
> are an expert... well

Is it really such a burden for "experts" to type '-f' occasionally
when doing strange configurations?

The problem with this in general is one of unpredictability.  You may
want to disable this check by default, but do you really want to
disable *all* checks?  That doesn't sound like a good idea... but then
if not, which checks *are* disabled?  It's not clear from the name,
and it's likely that we'll want to add or remove things, which means
having safety checks suddenly disappear... it's a UI nightmare.

I suppose I could accept a "force_all" option.  You'd have to be
basically daft to use it outside a testing environment, but it would
certainly be simple to define.  I still mostly think typing '-f' is a
better option.

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel