[docs] sensors docs: clarify run_key behavior in a single sensor evaluation

[Gw1p]

Summary & Motivation

I was caught off-guard in a production environment when a sensor that yields multiple RunRequests with the same run_key didn't de-duplicate them. Based on the current phrasing of the documentation, I would've thought that it would possibly take the first RunRequest and only launch that one but that doesn't seem to be the case.

I think it's better for the documentation to explicitly state what is the expected behavior in this scenario (assuming the current behavior is intentional).

How I Tested These Changes

Straightforward documentation change. Ran cd docs; cd next; yarn dev to spin up the docs locally and checked that it looks fine.

Changelog

Specified run_key behavior when emitting multiple RunRequests with the same run_key within a single sensor evaluation

[neverett]

I chatted with our engineering team and it sounds like this might not be supported behavior /cc @danielgafni

[Gw1p]

I chatted with our engineering team and it sounds like this might not be supported behavior /cc @danielgafni

Not supported in a sense that you're saying it doesn't currently work that way or in a sense that it shouldn't? If it's the former, happy to provide a minimum example. If the latter, happy to close the PR and file a bug instead.

An example would look something like:

@op
def dummy_op(context):
    context.log.info("Hello")
    yield Output(None)


@job
def dummy_job():
    dummy_op()


@sensor(
    job=dummy_job
)
def dummy_sensor(context):
    for i in range(3):
        yield RunRequest(
            run_key="5",
        )

On the first tick evaluation, it will launch 3 dummy_jobs, each with run_key of 5.

[neverett]

Not supported in a sense that you're saying it doesn't currently work that way or in a sense that it shouldn't? If it's the former, happy to provide a minimum example. If the latter, happy to close the PR and file a bug instead.

In the sense that it shouldn't, so feel free to file a bug, thanks!

[danielgafni]

I think this ping was supposed to be for @gibsondan?

[neverett]

I think this ping was supposed to be for @gibsondan?

Yes, my apologies!

[Gw1p]

Not supported in a sense that you're saying it doesn't currently work that way or in a sense that it shouldn't? If it's the former, happy to provide a minimum example. If the latter, happy to close the PR and file a bug instead.

In the sense that it shouldn't, so feel free to file a bug, thanks!

Gotcha, will do! I'm away for Christmas holidays until Thursday; will get a good example for it and file a bug once I'm back. Happy to look into fixing it as well.

[Gw1p]

Opened issue here: #26753

Closing the PR