Re: Bmcweb redfish multiple host support

[Ed Tanous <ed@tanous.net>]

On Tue, Jul 6, 2021 at 10:00 PM Naveen Moses S <naveenmoses@fb.com> wrote:
>
> Hi Ed,
>
> Thanks for your insights. we will work on it accordingly.
>
> If possible could you also clarify  the following :
>
> In redfish, /redfish/v1/Systems/system is used to store host/server related information.
> For LogServices, journal based logs are stored under the below path.
> /redfish/v1/Systems/system/LogServices/EventLog/Entries
> But few logs like sensor & FRU logs are related to BMC, which is displaying under system path. Could you please clarify why these logs are stored under this system path.

Because this path was analogous to the SEL log, and that part didn't
really matter at the time when there was only a single host;  It
likely could be improved, although maintaining backward compatibility
is going to be difficult.

> Whether can we take this under different path.

I'd need more details on what you're thinking here before I can render
an opinion.

>
> Thanks
>
> On 01-Jul-2021, at 9:58 PM, Ed Tanous <ed@tanous.net> wrote:
>
> On Thu, Jul 1, 2021 at 9:10 AM Naveen Moses S <naveenmoses@fb.com> wrote:
>
>
> Hello All,
>
> I need your suggestion on implementing support for  multiple computer systems or hosts under redfish implementation of BMCWEB.
>
> Redfish schema supports multiple computer systems under  redfish path "/redfish/v1/Systems/"
>
> Currently in bmcweb redfish core Computer systems path are is configured for as singlehost
>
> The redfish path computer system is given below
> "/redfish/v1/Systems/system”
>
> In case of multiple hosts can we use the following format.
>
> "/redfish/v1/Systems/system1”
> "/redfish/v1/Systems/system2”
> "/redfish/v1/Systems/system3”
> "/redfish/v1/Systems/system4”
>
> these paths should be populated based on the number of available hosts/computer systems
>
> Possible options to identify the host count are as given below:
> 1.pass the OBMC_HOST_INSTANCES as a macro definition to bmcweb
>
>
> Please don't do this.  In general we only use configure options for
> things that cannot be dynamically discovered on dbus or via the
> mapper, or where different implementations of things exist that meet
> the same API (like log service).  I don't think multi-host falls into
> either of those categories, as it can be seen on dbus.
>
> 2.Query dbus objects which has individual instance available for each host using object mapper.
>
>
> This would far and above be my preference, as it's similar to how the
> rest of the bmcweb URIs function.
>
>
> Is there any other alternate way available for identifying host instances at runtime under bmcweb?
>
>
> if we use the above format then can we change the computer system for single host
>
> as given below
>
> "/redfish/v1/Systems/system0”
>
>
> I would much prefer these are just based on the dbus path names
> instead of trying to explicitly pick a name.  That keeps bmcweb
> consistent with other resources, and systems that only implement a
> single host can choose to keep the same odata.id (system) if they
> want.
>
>
> Code reference : https://github.com/openbmc/bmcweb/blob/master/redfish-core/lib/systems.hpp
>
> method : requestRoutesSystemsCollection
>
>
>
> Thanks
>
>
> Looking forward to seeing the patches.  Thanks for looking into this.
>
> -Ed
>
>