Thanks for your response, and sorry for my very delayed response. I’ve answered inline below.
···
On Sun, Sep 4, 2016 at 10:10 PM, Morten Olav Hansen morten@dhis2.org wrote:
Hi Lorill
Sorry for the delay in answering.
Thanks, that is helpful.
We need to write our response parser to be able to parse any kind of error that may happen. Our program ingests CSV files from the user and we attempt to do as much error checking in advance as possible, but there may be other errors that happen when posting to the API such as validation errors, reference errors, or even 500 errors if there is an error within DHIS 2 itself. Given that we need to cater to every possible scenario, I’m not sure if we should be using “atomicMode=NONE” or not. Also, should there be “objectReports” regardless?
Agreed that using “lastImported” for /api/metadata is not very useful unless sending one single object. However, we are using our generic response parser for all DHIS 2 API endpoints and are trying to avoid the need for special case logic dependent upon which specific endpoint was called. Additionally, we do still use this field for many of the API calls where we do send one single object. Is the plan to deprecate the “lastImported” field across the API, or only in the /api/metadata call, so we can adjust our parser accordingly?
Although I understand that this might add more data to the payload, it would decrease the amount of data that we need to pass around in our app, and also potentially decrease the number of API calls we need to do. Not having this information in the response now means we need to pass around our original post data (which can be quite large, and which we aren’t currently doing) and do a lookup by ID to send any kind of useful information back to the user. Additionally, we may have even posted these objects by ID only, which means we will now need to do another API call to look up the object to get useful information to send back to the user. Keep in mind that in our app, the user doesn’t actually even have a knowledge of IDs - we are doing those lookups for them by name or code.
Would it be possible to add a flag to the api call where we can request the name or shortname or code to be returned in addition to just the ID? Perhaps similar to the metadata field filter? Where we could say supply importConflictsFields=:identifiable?
–
Lorill Crees
Project Leader / Senior Developer
2Paths Solutions Ltd.
lcrees@2paths.com
skype: lorill2paths
(604) 689-4123 x 15
- Is there any documentation outlining the response structure from all API calls, including when there are errors or any other scenarios? We have been upgrading our response parser on a trial and error basis attempting to mimic all possible scenarios which is not ideal.
Right now there is no documentation for this, but it should be easy enough to deal with I hope.
The basic structure is like this:
{
status: OK, ERROR, WARNING,
typeReports: [
{
klass: “Klass type”,
stats: {created, updated, deleted, ignored, total},
objectReports: [
{
klass: “Klass type”,
index: 123 (index into array you were sending),
uid: “uid of object”,
errorReports: [
{
errorCode: (see attached image),
message: “general message about what went wrong”,
mainKlass: (basically same klass as object report),
errorKlass: (class where it went wrong… if missing data element ref, this is DataElement),
value: (can be anything… something useful for error report to give more info)
}
]
}
]
}
]
}
- When submitting an identical post containing {“programTrackedEntityAttributes” […], “program” […]}, where there is an error in the Program being posted, the call to /api/metadata responds with success for “programTrackedEntityAttributes” with the correct “importCount” counts, but the call to /api/23/metadata responds for “programTrackedEntityAttributes” with “stats” counts containing all zeros, and no “objectReports”. It is like the first item being posted is completely ignored if the second item posted has errors. Is this the intended behaviour? And should there be “objectReports” regardless? Please see attached examples.
Yes, what kind of error are you getting? a reference error, or a validation error? in the old importer, it would always try to just import what it could… which was not always ideal (missing references etc was just ignored). If it’s just reference error, and you want to ignore them… please try with atomicMode=NONE, regarding the object report…
- There is no longer a “lastImported” field. Is this now deprecated?
There should now be a id exported in the object report… I think there might be some inconsistencies there though… I will go through it, in general this field was mostly used for our /api/type/id endpoints… were you also using it for /api/metadata? (it’s not very useful unless you are only sending 1 single object)
- The /api/23/metadata response with “errorReports” within “objectReports” appears to have replaced the /api/metadata “importConflicts”. In /api/metadata, the “importConflicts” “object” field contains the actual name of the object that has the error (eg: the Program short name in our case). However, in /api/23/metadata there is only the index which makes it more work to debug. Is it possible to also include the name of the object in the error report like the prior API version does?
Hm, right. I understand that it’s useful, but that can easily add up to a lot of data returned. Maybe we should add some kind of “debug” mode where more output is provided.
–
Morten
Thanks,
Lorill
Mailing list: https://launchpad.net/~dhis2-devs
Post to : dhis2-devs@lists.launchpad.net
Unsubscribe : https://launchpad.net/~dhis2-devs
More help : https://help.launchpad.net/ListHelp
