Why do we use `job_ids` as `Query` and not as `Body`?

[aldbr]

• diracx/diracx-routers/src/diracx/routers/jobs/status.py

  Line 40 in 21e62d9

  job_ids: Annotated[list[int], Query()],

• diracx/diracx-routers/src/diracx/routers/jobs/status.py

  Line 134 in 21e62d9

  job_ids: Annotated[list[int], Query()],

• diracx/diracx-routers/src/diracx/routers/jobs/sandboxes.py

  Line 169 in 21e62d9

  jobs_ids: Annotated[list[int], Query()],

This generates queries that can quickly become very long and very limited by the URL max length:

https://diracx.app/api/jobs/?job_ids=0&job_ids=12&job_ids=...

Trying to delete 1,000 elements in a single call generates an error for instance.
@TheauW bumps into the issue while making tests with diracx-web.

Is there any particular reason not to use the body here?

[chrisburr]

I think this came from a misguided effort to be "REST"-ful. I agree with changing it 😄

[TheauW]

I've started to change the endpoint, but in the tests we can't send a body with a DELETE request directly using FastAPI's TestClient — see fastapi/fastapi#5649.

So, the endpoint itself works, but we can't properly test it.

There are two possible solutions:

• Change the DELETE /jobs/ endpoint to POST /jobs/delete/ and pass the job IDs in the request body.
• Use client.request("DELETE", url, json={...}) in the tests to manually send a body — but this is more of a workaround and not a best practice.

Do you have any suggestions or preferences?

[aldbr]

I would go with POST /jobs/delete with the job IDs in the request body:

• because we are not following RESTful principles anyway
• because from what I can read, even if the RFC does not provide explicit instructions, having a body in a DELETE request is not widely supported (it makes sense from a RESTful point of view).

This can be discussed in the Bild Meeting.

Now I have a deeper question: why did we decide not to be RESTful?
I have a few ideas of course: it's more efficient to perform bulk operations. But do we actually need to perform bulk operations from the API?
I understand that internally we clearly want to perform bulk operations, through diracx-tasks for instance, but do users need to perform bulk operations?.
I also understand that external tools might need to perform bulk operations through the API, but should they? May be they should be integrated to the extensions for instance?

I think it would be great to take some notes about our design decisions, given the constraints that we have and then create "Architecture Decision Recods (ADR)" (in /docs/developers?).
As we will have people coming and leaving, it can potentially help them clearly understand our design choices and avoid them thinking too much about them without getting the full picture (as I am currently doing right now) or at least to do it on purpose.

https://medium.com/@nolomokgosi/basics-of-architecture-decision-records-adr-e09e00c636c6