Conformance classes instead of endpoint list

[m-mohr]

I'm wondering whether we can "easily" align better with OGC APIs by deprecating the endpoints entry in GET / and replace it with conformance classes in the long-term (2.0). In the short term (1.x) we would deprecate the existing functionality and add the new one.

I could imaging it working like this: Instead of specifying the method and path, you use the operation ID specified in the OpenAPI document.

For this I'm wondering, what the endpoint list is currently used for and by which software component.

• JS: Used primarily in the JS client, partially in the Web Editor and the Hub. Migration is relatively easy with the new proposal.
• Python: ? @soxofaan
• R: ?
• Back-ends: ?
• Aggregator: ? @soxofaan
• ...?

Example

Before

{
  "conformsTo": [
    "https://api.openeo.org/1.2.0",
    "https://api.openeo.org/extensions/commercial-data/0.1.0",
    "https://api.openeo.org/extensions/federation/0.1.0",
    "https://api.stacspec.org/v1.0.0/collections"
  ],
  "endpoints": [
    {
      "path": "/collections",
      "methods": [
        "GET"
      ]
    },
    {
      "path": "/collections/{collection_id}",
      "methods": [
        "GET"
      ]
    },
    {
      "path": "/processes",
      "methods": [
        "GET"
      ]
    },
    {
      "path": "/jobs",
      "methods": [
        "GET",
        "POST"
      ]
    },
    {
      "path": "/jobs/{job_id}",
      "methods": [
        "GET",
        "DELETE",
        "PATCH"
      ]
    },
    {
      "path": "/credentials/basic",
      "methods": [
        "GET"
      ]
    }
  ]
}

After

{
  "conformsTo": [
    "https://api.openeo.org/1.2.0",
    "https://api.openeo.org/1.2.0/operations/list-collections",
    "https://api.openeo.org/1.2.0/operations/describe-collection",
    "https://api.openeo.org/1.2.0/operations/list-processes",
    "https://api.openeo.org/1.2.0/operations/list-jobs",
    "https://api.openeo.org/1.2.0/operations/create-job",
    "https://api.openeo.org/1.2.0/operations/describe-job",
    "https://api.openeo.org/1.2.0/operations/update-job",
    "https://api.openeo.org/1.2.0/operations/delete-job",
    "https://api.openeo.org/1.2.0/operations/authenticate-basic",
    "https://api.openeo.org/extensions/commercial-data/0.1.0",
    "https://api.openeo.org/extensions/federation/0.1.0",
    "https://api.stacspec.org/v1.0.0/collections"
  ]
}

Thoughts?

[soxofaan]

In python client, there are two explicit cases of checking the endpoints:

• Connection.authenticate_basic will check if GET /credentials/basic is supported. Introduced by https://github.com/Open-EO/openeo-python-client/issues/247
• UDP usage (listing UDPs, saving a UDP) will check for /process_graphs support . Added in context of addressing https://github.com/Open-EO/openeo-aggregator/issues/90

Apart from that, there is also a user facing API to list and check endpoints (e.g. RESTCapabilities.list_features()), But it dates back from very early in the project (2018), I haven't it seen being used, I actually just found out by researching it for this issue, and it is not listed in the official API docs. I think we can safely consider this being completely unused.

[soxofaan]

In the aggregator:

• there is a check if GET /service_types is supported before trying to list service types of a certain backend, in order to merge everything across the federation

The aggregator generates its own endpoint listing independently from the endpoint listings of upstream backends.

Apart from that I couldn't find anything else that could be relevant here

[soxofaan]

In general, I kind of like the old style with explicit "path": "...", "methods": [ ...] listings as it is fully self-contained. With the "conformsTo" URLs, the client is kind of required to request and parse a whole bunch of additional URLs to get the same information.

[m-mohr]

I think we'll need to adopt this at some point for alignment with OGC APIs and deprecate our endpoint list.

We can probably simplify it, so that it's easier to translate endpoints to URLs 1:1 (i.e. it uses a template https://api.openeo.org/{apiVersion}/endpoints/{httpMethod}/{endpoint})

So for example:

{
  "conformsTo": [
    "https://api.openeo.org/1.2.0",
    "https://api.openeo.org/1.2.0/endpoints/get/collections",
    "https://api.openeo.org/1.2.0/endpoints/get/collections/{collection_id}",
    "https://api.openeo.org/1.2.0/endpoints/get/processes",
    "https://api.openeo.org/1.2.0/endpoints/get/jobs",
    "https://api.openeo.org/1.2.0/endpoints/post/jobs",
    "https://api.openeo.org/1.2.0/endpoints/get/jobs/{job_id}",
    "https://api.openeo.org/1.2.0/v/delete/jobs/{job_id}",
    "https://api.openeo.org/1.2.0/endpoints/patch/jobs/{job_id}",
    "https://api.openeo.org/1.2.0/endpoints/get/credentials/basic",
    "https://api.openeo.org/extensions/commercial-data/0.1.0",
    "https://api.openeo.org/extensions/federation/0.1.0",
    "https://api.stacspec.org/v1.0.0/collections"
  ]
}

so for example if you have supports(method: str, path: str) : bool you could simply so something like:

def supports(method: str, path: str) -> bool:
  escaped_path = path.replace("{", "\{").replace("}", "\}")
  method = method.lower()
  pattern = f"https://api.openeo.org/[^/]+/{method}/{escaped_path}"
  for c in conformsTo:
    if c.match(pattern):
      return True
  return False

supports("GET", "/collections")

[soxofaan]

I'm fine with this if it's acceptable to just (naively) parse the listed conformsTo URLs, without having to actually request them and inspect their responses

[m-mohr]

Sure, conformance classes often don't resovle in OGC land anyway. They will likely just give 404s or resolve to something that's not overly useful...