maas-devel team mailing list archive

[Julian Edwards <julian.edwards@xxxxxxxxxxxxx>]

On 22/01/14 23:20, Gavin Panella wrote:
> On 21 January 2014 13:52, Andres Rodriguez <andreserl@xxxxxxxxxxxxx> wrote:
>> All,
>>>
>>> Pondering on what Jeff says though, should a node even get to the
>>> Ready state without a power type?  Any chance we can require a power
>>> type at the point of accepting a node?
>>
>> I personally believe that not allowing a machine to get to the Ready
>> state without a power type is an inappropriate solution.
> 
> Can you explain why? I think it would be quite good to hold back nodes
> until the means of power control have been specified.

For manual operation.  I think this boils down to internal
representation of it, but the empty power type means manual.

> I think there should not be a default power type. If it cannot be
> discovered, the node is not "ready" and intervention is required.

I agree, that's why I just removed the default power type.  If nobody
set the power type, it remains empty and effectively "manual".

Whether we want to allow this is another matter - should we ask the user
to explicitly say it's manual?  Quite probably.

> If we need to support manual power control, that ought to be an option.
> However, this opens a can of worms: how do we tell a user when we need
> them to power a machine on or off? In the UI we can add a message. Via
> the API we would need a new way to indicate this to caller of start().
> All clients would need to grok this, including Juju and the Juju GUI.

We don't really want to support manual control in production scenarios,
I see this as a testing facility.

We could have a setting that says whether to allow manual power types or
not, which means the user has to make a conscious decision (again).

> This is not specific to power control: it's a problem with warnings in
> general. They're hard to get in front of a user, and harder still to get
> a user to heed them.

Yup.

> However... with the new driver model we're aiming for, each site could
> customise their own "Manual" (or "Externally-managed") driver. One site
> might send a text message, another might activate a solenoid glued to a
> machine's switch. My point is that it would no longer be MAAS's problem;
> the user has explicity told MAAS that it's his/her problem.

Yes - although I don't think we should encourage manual power types
outside of a testing environment.


J