RFC: new design of phosphor-time-manager on sdbusplus

[vishwa]

On 01/18/2017 08:14 PM, Patrick Williams wrote:
> On Tue, Jan 17, 2017 at 03:51:39PM +0800, Mine wrote:
>> On Tue, Jan 17, 2017 at 3:44 AM, Patrick Williams <patrick at stwcx.xyz> wrote:
>>> On Fri, Jan 13, 2017 at 03:42:50PM +0800, Mine wrote:
>> I should be meant to create two "instances" on the bus, eventually it looks
>> something like:
>> ```
>> xyz.openbmc_project.Time.Manager
>>      /xyz/openbmc_project/time/host_epoch
>>          org.freedesktop.DBus.Peer
>>          org.freedesktop.DBus.Introspectable
>>          org.freedesktop.DBus.Properties
>>          xyz.openbmc_project.Time.EpochTime
>>      /xyz/openbmc_project/time/bmc_epoch
>>          org.freedesktop.DBus.Peer
>>          org.freedesktop.DBus.Introspectable
>>          org.freedesktop.DBus.Properties
>>          xyz.openbmc_project.Time.EpochTime
> There isn't a reason to name the object path "epoch".  /time/bmc and
> /time/host0 is probably more appropriate.
>
>>>> And there will be no “curr_time_mode/owner” or “requested_time_mode/owner”
>>>> properties on DBUS.
>>> So these are only stored in phosphor-settingsd now or are they also used
>>> internally for decisions?  I believe the previous implementation had
>>> them exposed more for debug purposes.  Are you going to add them to the
>>> journal at least?
>> Yes, the time_mode and time_owner are still stored in
>> phosphor-settingsd, and they
>> are used internally by phosphor-time-manager, which will register
>> callback for the
>> settings' change and handled accordingly.
>> The difference from the previous implementation is that we do not expose these
>> settings in time-manager's DBus now.
> There are two aspects for consideration here:
>
>      1. The current time manager defers changing the host policies until
>         the next boot.  We need to continue this behavior.
There was a concern from test team on 'why do we need to defer applying 
any settings that are used by time manager'.

I thought and partially agree to their concern. For changing the owner 
*away from SPLIT*, we can actually consume that readily. Its only the 
change * TO SPLIT* that needs to delayed until off.

Also, the time_mode changes can be applied readily ( although time 
manager may not like HOST and NTP and in that case, its in pending state 
anyway ).


>
>      2. If the process restarts we need it to go back into the "current
>         state" and not the "requested state".  How do we make this
>         happen?  The current implementation might also have a flaw here
>         so maybe we log it as an issue for follow up.


The current implementation only sets the values based on what was set 
prior to reset.
It does that by consuming the saved data in /var/lib/obmc/saved_*. So 
there is not an issue there.


>
>> What do you mean by "add them to the journal"?
> Use phosphor-logging.
>
>
>
> _______________________________________________
> openbmc mailing list
> openbmc at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/openbmc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/openbmc/attachments/20170119/fa3ce1d5/attachment.html>