Feature request in phosphor-pid-control

[Patrick Venture]

It follows a similar downstream feature where a sensor being
"unavailable" because the mapper is swamped shouldn't except the
program.  The only difference here is that this is a sensor which also
needs to be unable to throw the system into failsafe mode.  The system
starts in failsafe mode until it's heard from all the sensors it's
expecting.  This would need to check to see if that sensor supports
the notion of a failsafe timeout -- which is already supported.

A sensor triggers failsafe IFF the value is older than the timeout
value and the timeout value is non-zero.  So, in theory, the
initialization loop could check that the timeout value for a sensor is
non-zero before adding it to the fail-safe check-in list.

So, this change really is just, "don't fail if you can't find the
sensor during start-up, and keep trying until it shows up"

Does that follow?

Patrick

On Thu, Jun 14, 2018 at 12:24 AM, Lei YU <mine260309 at gmail.com> wrote:
> This email is about a feature request in phosphor-pid-control, that to
> support
> optional sensors.
> It is mentioned in comment of https://gerrit.openbmc-project.xyz/#/c/11003/
>
> Background:
> Current phosphor-pid-control checks the temperature sensors and calculate
> the
> expected fan speed.
> It assumes the temperature sensors are always available. And if the sensor
> is
> not there, the service throws exceptions.
>
> The feature request:
> I would like phosphor-pid-control to support "optional" sensors, that it may
> be unavailable when system is running.
> Such sensor is counted when it is available, and is ignored when it is
> missing.
>
> An use case:
> I want to adjust fan speed based on cpu core or dimm temperature.
> For a typical OpenPOWER P9 system, there could be at most 48 cores, but they
> can be missing depending on the CPU, e.g. if a system is plugged CPUs with
> 22
> cores, 4 cores temperature will be missing. And in run time, a core may be
> garded if it has problem it will be missing as well.
> The same for dimms, for a system with at most 16 dimms, it depends on how
> many
> dimms are attached.
> In such case, the config will specify all the temperatures for cpu cores and
> dimms, and phosphor-pid-control shall ignore the missing temperatures.
>
> Patrick, what do you think about such case?
>
> Thanks!