The type of sensor value in redfish should be double or int64?

[Brad Bishop]

at 7:19 PM, James Feist <james.feist at linux.intel.com> wrote:

> On 10/15/19 4:15 PM, Brad Bishop wrote:
>> at 6:19 PM, Ed Tanous <ed.tanous at intel.com> wrote:
>>> On 10/15/19 12:04 PM, Brad Bishop wrote:
>>>> Is it possible to get bmcweb supporting patch with double and int?  On
>>>> the surface this seems like the shortest path to enabling Carol.
>>>
>>> Unless phosphor-hwmon is patched to support settable sensors, I don't
>>> think there's a good reason to add support to bmcweb.
>> Ok.  A couple different questions then.
>> What does it mean to be settable?  phosphor-hwmon implements  
>> Sensor.Value and, (in violation of the interface documentation), doesn’t  
>> prevent an external application from writing to the properties in it.
>> I assumed the assertion that phosphor-hwmon doesn’t provide settable  
>> sensors is because the value set by an external application doesn’t  
>> stick?  The phosphor-hwmon application logic will write over it the next  
>> time it polls the hwmon device.
>> So what does dbus-sensors do for hardware sensors?  Does the application  
>> ignore the hardware state after being patched via Redfish?
>
> Yes until the sensor is reset.

Ok.  Well this behavior could be trivially added to phosphor-hwmon too if  
that improves the case for adding patch support for the integer interface.

>
>
>>> With that said,
>>> it's just software, it could certainly be done.  Someone would need to
>>> sort out how to manage the scale factor, as a Redfish PATCH request
>>> doesn't have a concept of scale, and operates on a float, so you'd have
>>> to do some software handholding to convert the value to something
>>> appropriate before it was scaled to the int64.
>>>
>>> With that said, all of this seems like a lot of work.
>> I don’t know of any alternatives that are less work.  If there are some,  
>> I’d love to hear about them.  There were several years of OpenBMC  
>> development against the integer interface signature before double became  
>> a problem - as Patrick mentioned - undoing all that will not be trivial.
>>> Just to clarify in case anyone isn't clear, the /xyz/* and /org/* type
>>> REST handlers do support SetProperty for int and double, and introspects
>>> the daemon to determine which one should be used.  So far as I'm aware,
>>> the only thing non-functional is a Redfish patch of a sensor, as it's a
>>> bit more complex to do properly.
>> I had a look at the Sensor schema in the latest Resource and Schema  
>> guide and the sensor reading property is read only.  Why do we even  
>> allow it to be patched?
>
> The intention was for this to only be in manufacturing test mode (not  
> sure if that is true today or not). So under a normal user, it would be  
> read only.

Ok, makes sense.  thx for the info James.

>
>>> -Ed