[Nmc-wg] Mappings: old result codes -> new result codes

[romradz at man.poznan.pl]

Hi Jason,

On Fri, 1 Oct 2010, Jason Zurawski wrote:

> Hi Roman;
>
> Here are some additional comments to back up what we discussed on the call:
>
> On 9/30/10 8:45 AM, Roman Łapacz wrote:
>> Hi Jason, Aaron and others,
>> 
>> I attached mapping text file and MS Word doc with small updates (just to
>> have them together in one email).
>> 
>> W dniu 2010-05-27 21:08, Jason Zurawski pisze:
>>> Hi Roman/All;
>>> 
>>> Some comments on the proposal:
>>> 
>>> "Wrong structure of a request":
>>> 
>>> Like Aaron, I think I am having a hard time with some of these as
>>> being a purely 'structural' issue. For instance
>>> 'error.ls.data_trigger' - I would assert there is some context to be
>>> known about the content of the rest of the message before calling this
>>> structural issue (for instance if there was a data/metadata pair
>>> already).
>> 
>> If there is such data/metadata pair in a message than I wouldn't expect
>> to get error.ls.data_trigger. An example of xml message would help to
>> analyse such case.
>
>
> So the example I had in mind was something like this:
>
> <message>
>  <metadata id="m1">
>    <!-- ... something ... -->
>  </metadat>
>  <data id="d1" metadataIdRef="m1" />
>
>  <metadata id="m2">
>    <!-- ... something ... -->
>  </metadat>
>
>  <!-- no trigger for m2, and m2 is not related to m1 or d1 -->
> </message>
>
> In this case m1 and d1 could be acted on, and m2 could not (it would get the 
> error.ls_data_trigger response).  The pSPS would return the results for the 
> first metadata/data pair and an error for the second in the same message.

Geant MA services with base1 library were able to accept such messages. 
Services with base2 probably are able as well but I don't remember it in 
100%.
In general for me a message with more than one data triggers contains 
logical separate messages which should be processed independently.

>
> My objection is related to the proposed 
> "http://perfsonar.net/status/clienterror/wrong_message_structure/" eventType. 
> In this case the message structure is syntactically correct, the semantics 
> are just wrong in one instance.

First chain in the example is correct and the response should contain some 
data. The metadata m2 is hanging and it may be difficult to figure out by 
a service is it a part of functional chaining (so somehow connected 
with metadata m1) or is data trigger d2 missing. Proposed event type name 
is incorrect but more general one, e.g.
http://perfsonar.net/status/clienterror/wrong_structure/
may be suitable.


<message>
   <metadata id="m1">
     <!-- ... -->
   </metadata>
   <data id="d1" metadataIdRef="m1">
     <!-- ... -->
   </data>
   <metadata id="m2">
     <!-- http://perfsonar.net/status/clienterror/wrong_structure/  -->
     <!-- for hanging m2 -->
   </metadata>
   <data id="d2" metadataIdRef="m2">
     <!-- datum with some explanation  -->
   </data> 
</message>



Even we could ignore m2 because it does not belong to any chain in the 
message.


>
>
>>> I think most of these are right, but we should be careful with the
>>> context before calling all of them a pure xml structure violation.
>> 
>> I still have the problem how much context should be included in the code
>> (mainly datum element contains a description).
>
>
> I forget what the outcome of the early discussion was (I am sure someone can 
> correct me), but I believe that the description that will be returned in the 
> datum does not need to be a 'standard' message.  The eventType code and 
> instructions guiding its use (e.g. 'use this this code for errors that are 
> related to ...') are the only thing we should make standard.  If someone 
> wants to send back a datum that contains more than what the standard 
> definition allows (e.g. a stack trace perhaps, or error logs) this seems fine 
> to me.

Agree.

>
>
>>> 
>>> "EventType in a request is not supported":
>>> 
>>> This also seems more context sensitive to me.
>>> 
>>> "Request is not supported":
>>> 
>>> I think 'message type unsupported' is different than 'no message type
>>> specified'
>> 
>> It just says ther's a problem with message type. Can we assume that if
>> there's no message type then default type is considered (default type
>> may be unspecified).
>
>
> I dont think we should be assuming a default message type, because what would 
> be the default?  Services like the LS use different messages than the MAs and 
> MPs.  I would still favor having both 'type unsupported' and 'no type 
> specified'.

OK. That sounds reasonable.
(I was thinking that each service could have such default message type 
depending on service type but that would fail if an implementation 
represented two service types)

>
>
>>> 
>>> "Elements of a request are not supported"
>>> 
>>> I think we may need to be a bit more clear in the message. Instead of
>>> 'wrong XXX' we may want to say 'unexpected XXX element' or something
>>> similar. This will clear up of the content is wrong or the element is
>>> wrong.
>> 
>> Doesn't message_element_not_supported match here?
>
>
> Good question.  I guess I can see this happening in one of two ways:
>
> 1) the element is unexpected semantically, but still schematically legal. 
> For example:
>
> <ns1:metadata>
>    <ns:1subject />
>    <ns1:eventType />
>    <ns1:parameters />
>    <ns2:parameters />
> </ns1:metadata>
>
> 2) The element is not semantically or schematically expected.  For example
>
> <ns1:metadata>
>    <ns:1subject />
>    <ns1:eventType />
>    <ns1:parameters />
>    <ns1:datum />
> </ns1:metadata>
>
> I would say these are different errors.  In #1 the parser may choose to 
> interpret the 'last' parameters element it saw, so things may still 'work'. 
> We may owe the user a warning to say that something didnt look right though. 
> In #2 the element is unexpected and may be no-recoverable from a parsing 
> standpoint.

So we have two groups of errors: semantic and structural. The HTTP style 
of grouping does not show it but maybe it would be useful.

Roman


>
> Hope this helps;
>
> -jason
>
>
>> Roman
>> 
>>> 
>>> Thanks;
>>> 
>>> -jason
>>> 
>>> 
>>> On 5/27/10 10:56 AM, romradz at man.poznan.pl wrote:
>>>> 
>>>> Hi Aron,
>>>> 
>>>> On Thu, 27 May 2010, Aaron Brown wrote:
>>>> 
>>>>> One minor thing. I'd move the "No message type specified" to the
>>>>> "wrong_message_structure" since the request is missing a required
>>>>> element.
>>>>> I'm curious what the "message_element_not_supported" event type is
>>>>> for. From the error messages listed, It wasn't obvious to me why the
>>>>> service threw
>>>>> them.
>>>> 
>>>> for example, if an element located in a request is not recognized (or
>>>> the content of xml tag is wrong, eg. ip address)
>>>> 
>>>> Roman
>>>> 
>>>> 
>>>>> 
>>>>> Cheers,
>>>>> Aaron
>>>>> 
>>>>> On May 27, 2010, at 10:32 AM, Roman Lapacz wrote:
>>>>> 
>>>>> 
>>>>> Hi,
>>>>> 
>>>>> I've started preparing the mappings. So far I've focused only on RRD
>>>>> MA, PSPS commons and PSPS LS. This is just the beginning for the
>>>>> discussion (to check if this is a good direction for all interested).
>>>>> 
>>>>> Cheers,
>>>>> Roman
>>>>> <proposal-of-mappings-20100527v1.txt>
>