wmo-im / wis2-notification-message

WIS 2.0 MQP message to notify users of availability of new data
https://wmo-im.github.io/wis2-notification-message
2 stars 2 forks source link

consider coarser resource publication with interactive links #41

Open tomkralidis opened 1 year ago

tomkralidis commented 1 year ago

A WNM canonical link must be opaque such that no further interaction is required for download.

At TT-WISMD 2023-04-12, @amilan17 discussed that ECMWF has a use case where a given notification can point to a canonical link that takes further interaction from the user to be able to download data (granules). An example here could be advertising a model run with many parameters/pressure levels/timesteps.

cc'ing @sebvi for clarification (not sure the above captures things clearly enough).

sebvi commented 1 year ago

The question I asked is the following: is it ok to put as a link our api rather than a link to the data directly? From your first sentence it does not seem to be desirable.

We typically produce daily data i huge files containing all parameters, steps, etc. and users might not want to download these huge files so we typically point them to our api where they can download only the portion they are interested in.

tomkralidis commented 1 year ago

Messages can certainly add additional links for APIs, as can discovery metadata. Users who are subscribed to ECMWF data, for example, can opt to use the API in lieu of the canonical link.

kaiwirt commented 1 year ago

I think if you just follow the link in the WNM you should receive something meaningful. This might be a default data set in which case the link could well take additional parameters allowing to download other data sets.

Other than that we could add another type of WNM (maybe on a different topic?) specifically for API endpoints which just informs users of updates. The way they need to query the data can then be defined in the metadata for example using OpenAPI and is not part of the WNM.

amilan17 commented 1 year ago

related to wmo-im/tt-nwpmd#13

petersilva commented 1 year ago

fwiw, feeds are about real-time data... processing the fraction of data available as soon as possible. NWP products for a run are likely not available at once. One would imagine publishing the products for analysis, then +6h, +12h, +18, etc... the availability would be separated by between 5-15 minutes. In addition, there may be derived/transformed products that are made available after the initial ones. If you group all the products under one announcement, can likely only announce at the end of each run, which may be upto 90 minutes after the data is available. It's not real-time anymore.

tomkralidis commented 1 year ago

TT-WISMD 2023-09-11:

Specification updates:

This is a similar workflow of EUMETSAT.

@sebvi / @gaubert can you provide an example that we can include in the WNM annex of examples?

tomkralidis commented 9 months ago

TT-WISMD 2024-01-15:

gaubert commented 9 months ago

@tomkralidis example provided in my pull request.

tomkralidis commented 9 months ago

Thanks @gaubert. @sebvi any chance you can also provide an example?

6a6d74 commented 9 months ago

Jeremy (@6a6d74) will incorporate the consensus above into the Guide; section for data publishers 2.6.3

6a6d74 commented 9 months ago

@tomkralidis - do you know which section of the WNM spec this material will go into? I'd like to reference it from the WIS2 Guide.

tomkralidis commented 9 months ago

@6a6d74 this would be put forth in https://wmo-im.github.io/wis2-notification-message/standard/wis2-notification-message-DRAFT.html#_links

josusky commented 9 months ago

I would like to emphasize that while one click to get data is nice for showcasing, it is practical only for small things like station observations. Those are a corner case in fact. Satellite, NWP and radar data are much more important and for those it is not very practical to provide a link to download global coverage with one click. Sure, in the NWP case mentioned by @petersilva it si desirable that the publisher announces data availability incrementally as the model produces individual granules, but the granules will have size in order of hundreds of megabytes. It would be very impractical to send one notification per grid point per parameter per level per forecast hour ... The subscribers of such services (and those services are the main use case) should be a bit more sophisticated than a simple "grab the URL". That is, the notifications about Satellite, NWP and radar data may have a "one click link" that points to an API/service description document (that is the README mentioned by @tomkralidis) and additional information that tells the (sophisticated) clients which forecast hour, parameter etc. has become available.