Redfish EventService Implementation

Wed May 27 21:50:26 AEST 2020

Apparao,

Thanks a lot for your suggestions. We lean towards using a dictionary to 
map Redfish ResourceType to D-Bus objects path and vice versa and then 
using D-Bus match to generate life cycle events. This way, the changes 
are limited to bmcweb. The resource type and the origin resource URI 
will be included in the event log. This requires change in the format of 
event log file /var/log/redfish. I have commented the same in the server 
sent event patch that you have uploaded. Kindly see if you can leave the 
parsing of file to the OEMs. That way, the existing infrastructure can 
be used by the OEMs to support other filtering mechanisms as defined in 
the specification.

Thanks,
Rajes

On 27-05-2020 09:18, Puli, Apparao wrote:
>
> Hi Rajeswaran,
>
>   Thanks for your mail. At the moment, I don't have plans to support 
> the "ResourceTypes", "OriginResources" based filtering. Basically 
> Intel uses file systems based redfish event logs( journalctl -> 
> rsync-> filesystem) and doesn't use D-Bus mechanism like IBM uses. So 
> I am not much familiar with D-Bus logging but some of the suggestions:
>
>  1) While logging redfish events over D-Bus itself,  it can provide 
> details on ResourceTypes and OriginResource URI/Path.
>
>      This is ideal and most efficient way. Since  we walked a walked 
> long distance from start, Its hard to modify all the services to uses 
> these 2 new input parameters while logging events( Requires change in 
> almost all repo's)
>
> 2) For resourcesTypes: Can have mapping dictionary against all 
> MessageId's. For OriginResources: I believe, event log over D-Bus is 
> already holding the Path. If so, last 3/4 nodes of uri can be taken 
> and mapped against the resources and that can be used in Event 
> filtering. We did used same mechanism in case of telemetry  while 
> mapping MetricReportDefinitions to URI.
>
> Hope this helps.
>
> Thanks,
>
> -Appu
>
>
> On 5/26/2020 5:50 PM, RAJESWARAN THILLAIGOVINDAN wrote:
>>
>> Apparao,
>>
>> I see that you have uploaded a 
>> patch(https://gerrit.openbmc-project.xyz/c/openbmc/bmcweb/+/32441) 
>> for supporting server sent events. This patch supports event 
>> filtering based on registry prefix  and/or messageId.
>>
>> I would like to know if you have any plan to support event filtering 
>> based on resource type. If so, I would like to work together for a 
>> better solution. Earlier, I have proposed a solution based on D-Bus 
>> match using a dictionary. With that approach, the major challenge is 
>> to map Redfish resource and properties to D-Bus object and properties 
>> respectively.   If D-Bus applications are modified to include 
>> resource type and origin of condition in the event, then there is no 
>> need for a map. But,that brings Redfish terminology to the 
>> application. Also, this will become an overhead if an OEM is not 
>> interested in Redfish event service.
>>
>> Thanks,
>> Rajes
>> On 25-02-2020 19:36, Puli, Apparao wrote:
>>>
>>> Hi Ratan,
>>>
>>>    Comments in-line
>>>
>>> On 2/24/2020 12:07 PM, Ratan Gupta wrote:
>>>>
>>>> Hi Apparao,
>>>>
>>>> On 2/20/20 12:49 AM, Puli, Apparao wrote:
>>>>> Hi,
>>>>>
>>>>>   I am sorry for late response as this mail is buried under and 
>>>>> got struck back of my mind.
>>>>>
>>>>> As i did mentioned in EventService design document, EventLog 
>>>>> Monitor service is not common across the organizations( Example: 
>>>>> Intel uses the redfish event logs file and inotify mechanism for 
>>>>> monitoring the event logs. Where as IBM uses d-bus event log 
>>>>> mechanism and they can relay on match rules). That said challenges 
>>>>> with ResourceType mapping will be different in both monitoring 
>>>>> mechanisms. This is good point. Initially when i started 
>>>>> EventService design, i thought we can have mapping in bmcweb for 
>>>>> ResourceTypes with set of MessageID's for Logged events ( As per 
>>>>> Intel design) but not sure that may become difficult when we 
>>>>> expand supported ResourceTypes.
>>>>
>>>> If I am getting correctly, Here is the flow which Intel uses.
>>>>
>>>>  1. Individual repo have to push the logs using sd_journal_send
>>>>     which will write to the file(/var/log/redfish) by using rsyslog
>>>>     daemon
>>>>
>>>> sd_journal_send("MESSAGE=%s","journal text","PRIORITY=%i", <LOG_LEVEL>,
>>>>                  "REDFISH_MESSAGE_ID=%s",
>>>>                  "ResourceEvent.1.0.ResourceCreated",NULL);
>>>>
>>>>       * How you would populate the "OriginOfCondition" during
>>>>         sending of event?
>>>>         (https://redfish.dmtf.org/schemas/v1/Event.v1_4_1.json)
>>>>
>>> Currently in logServices( logEntry),  we are not reporting the 
>>> "OriginOfCondition" property as per schema. I will check with Jason( 
>>> Who wrote the logService) and get back on this.
>>>
>>> BTW can you give me how this information is fetched in IBM way of 
>>> LogService implementation( D-Bus)? If you already ratified any such, 
>>> i think we can leverage.  If not, We work together for solution.
>>>
>>>>       * Any plan to add resource path in this journal message which
>>>>         tells that this is the path for which resource created
>>>>         event generated.
>>>>
>>> Same as above.
>>>>
>>>>       * Would the path be "Redfish Path/ D-bus Path"?
>>>>
>>> As per Redfish specification, This should be "@odata.id" which means 
>>> it should be of resource uri and so we can't use d-bus path here for 
>>> OriginOfConditions.
>>>>
>>>>       * Where the mapping would be done between D-busPath/Redfish
>>>>         Resource path?
>>>>
>>>>      
>>>>           Cons: Every application have to make the change(use sd_journal_send).
>>>>           My thought is backend application should not be aware of the redfish terminlogy.
>>>
>>> Having separate process only for this mapping may not be good( No 
>>> different from maintaining that map inside bmcweb as there won't be 
>>> any other consumers). Ideal way is, that should be mapped while 
>>> logging logEntry's itself. But we are not doing it currently which 
>>> need to be re-looked. Give me some time, I will think and check with 
>>> other folks and get back.
>>>
>>>> *2.* Some application(bmcweb) would do the Inotify on the 
>>>> path(/var/log/redfish) and send the event once there is any 
>>>> activity on this file.
>>>>
>>>> > I thought we can have mapping in bmcweb for ResourceTypes with set of MessageID's for Logged events ( As 
>>>> per Intel design)
>>>>
>>>>      Can you explain more here. What is your plan? How you would do the Resource Type based event filtering?REDFISH_MESSAGE_ID is different than the resource type.
>>> Initially i thought "ResourceType" based event filtering can be done 
>>> using minimal mapping( Using MessageID and args). But that will work 
>>> for minimal set. If the supported ResourceTypes grows, we will have 
>>> bigger challenges which i can sense it now.  Anyway, Supported 
>>> Resources are completely implementation specific. If this value is 
>>> empty means, by default all event logs will be sent to subscribers. 
>>> This is what we can start with before supported  ResourceTypes list 
>>> grows.
>>>>>
>>>>> As per my reading from below query, You are looking at d-bus match 
>>>>> rules and ResourceTypes mapping which is more specific to d-bus 
>>>>> event logging(IBM way of implementing event logging). reading it 
>>>>> from journal logs will give more information but that will impact 
>>>>> the performance to large extent. This might be one of the reason 
>>>>> why we (Intel) uses Redfish message ID while logging redfish 
>>>>> events logs to journal(You can refer design document for same at 
>>>>> https://github.com/openbmc/docs/blob/master/architecture/redfish-logging-in-bmcweb.md). 
>>>>> In opinion, in your d-bus if you are using some kind of 
>>>>> filter(Example REDFISH_MESSAGE_ID) while logging in journal logs 
>>>>> for all events and figure out the way to monitor the journal logs 
>>>>> without impacting the performance, that should be ok as long as 
>>>>> match filters are satisfied for Redfish EventService subscriptions 
>>>>> and supported Types(Again differs with implementation).
>>>>>
>>>>> Thanks,
>>>>>
>>>>> -Appu
>>>>>
>>>>> On 2/10/2020 1:52 AM, RAJESWARAN THILLAIGOVINDAN wrote:
>>>>>> ApparaRao.
>>>>>>
>>>>>> As you have shown interest in this feature and submitted the 
>>>>>> design document, do you have any opinion on this? Do you see any 
>>>>>> merit in using D-Bus match in bmcweb to create event logs for 
>>>>>> life cycle events?  Please feel free to weigh in.
>>>>>>
>>>>>> Thanks,
>>>>>> Rajes
>>>>>>
>>>>>> On 01-02-2020 02:23, RAJESWARAN THILLAIGOVINDAN wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> I am going through the bmcweb code for implementing Redfish 
>>>>>>> EventService based on the design document 
>>>>>>> https://gerrit.openbmc-project.xyz/c/openbmc/docs/+/24749. This 
>>>>>>> design is hooked to the journal based Redfish Event Logging. For 
>>>>>>> life cycle events(ResourceAdded, ResourceRemoved, 
>>>>>>> ResourceUpdated),  using D-Bus match, bmcweb can create an event 
>>>>>>> log. This requires a JSON dictionary, comprising an array of 
>>>>>>> Redfish Resource Name and the D-Bus path. This approach works 
>>>>>>> only in case of one to one mapping of Redfish Resource Name and 
>>>>>>> the D-Bus path. For propertiesChanged events, if the Redfish 
>>>>>>> Resource property is not on the same D-Bus path or the Redfish 
>>>>>>> Resource property name is different from the D-Bus property 
>>>>>>> name, then an additional JSON dictionary to maintain this 
>>>>>>> information is required. With D-Bus match alone in the bmcweb, 
>>>>>>> Redfish EventService can't be fully supported. For the Message 
>>>>>>> Registers and the Resource Types that are supported, the 
>>>>>>> relevant OpenBMC application must create an event log in the 
>>>>>>> journal using either the phosphor::logging::entry or 
>>>>>>> sd_journal_send() command.
>>>>>>>
>>>>>>> After realizing that with D-Bus match in the bmcweb alone can't 
>>>>>>> help to fully implement EventService, I prefer to avoid using 
>>>>>>> D-Bus match in bmcweb. Instead, I prefer to modify the OpenBMC 
>>>>>>> application that generated the event to create an event log in 
>>>>>>> the journal. Do you see any advantage of using combination of 
>>>>>>> D-Bus match in the bmcweb wherever it is possible and changes to 
>>>>>>> OpenBMC application in other cases to create an event log ?
>>>>>>>
>>>>>>> Your views are highly appreciated.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Rajes
>>>>>>
>>>> Thanks
>>>> Ratan
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/openbmc/attachments/20200527/8dd13296/attachment-0001.htm>