Add dispatchable workflow for deleting test model runs

[jeancochrane]

This PR adds a new workflow, delete-model-runs that can be dispatched with a comma-delimited string of model run IDs to delete models that have been run for the current assessment cycle.

Models in previous assessment cycles are protected from deletion in two ways:

• The script will derive the current assessment cycle year, and will only look for model runs with the given ID in that year; if the model run is not present in the current year, the script will raise an error before attempting to delete any runs
• The IAM role that the script is run with only has permissions to delete models for the current assessment year, and AWS will raise an error if the workflow somehow attempts to delete models from a previous year

One high level question about the condo model: I've been assuming that we need to port these changes over to the condo model before we can count #56 as complete, but taking a look at the file_dict.csvs for both repos, it seems like there aren't actually any differences in the output paths for the model artifacts; the models are simply distinguished by their autogenerated run IDs. Do we need to port this code over to the condo model, or is it good enough to use the workflow in this repo to delete condo model runs? I think the tradeoff is that it would be more confusing from an operational perspective to need to delete condo model runs from the context of a model-res-avm workflow, but at the same time it would make long-term maintenance easier since we don't need to maintain two copies of the delete-model-runs workflow (I'm assuming that it's not worth extracting this logic out into a shared external workflow as we did for build-and-run-model because I don't imagine that this logic will be particularly useful outside the context of the res and condo model repos).

Connects #56.

[jeancochrane]

I agonized over this interface, perhaps unnecessarily; I'm curious if you agree that a single comma-delimited string is the right compromise!

[dfsnow]

I think it's fine! Though we might want to use a space-delimited string in order to match build-and-test-dbt. It would have to be quoted i.e. not positional args. I'm really fine with either though as long as it's clear what the input is supposed to be.

[jeancochrane]

Cool, I switched the workflow input to accept a space-delimited string in a4895c2.

[jeancochrane]

Does this heuristic seem reasonable to you?

[dfsnow]

Totally! I would bump the cutoff month to May though, as sometimes we still do model touchup stuff in April.

[jeancochrane]

Done in a4895c2!

[jeancochrane]

To test deletion, I copied over run artifacts for two different model runs from year=2023/ to year=2024/ and deleted the 2024 versions. Example workflow here: https://github.com/ccao-data/model-res-avm/actions/runs/6908697011/job/18798517688

In order to delete existing test model runs from year=2023/, my thinking was that we could make a list of model runs we want to delete, and then I can move them from year=2023/ to year=2024/ and delete them. I'll need your signoff on this list to make sure I don't delete anything important, and I'll plan to do this task after merging this PR.

[dfsnow]

That works for me! Let's also delete other test runs created during the model infra construction.

[jeancochrane]

Sounds good, will do!

[dfsnow]

@jeancochrane Just a quick answer to the PR body before a full review: no, we don't need to port this over to the condo model, since it shares the exact same run/output structure. We can just use this workflow to delete runs from both models, even if it's a bit confusing.

[dfsnow]

Looks great @jeancochrane. This will be quite handy once we start flooding S3 with useless test models :D

[dfsnow]

nitpick: Can't you just pass the input directly here?

Suggested change

	run: Rscript ./R/delete_current_year_model_runs.R "$RUN_IDS"
	shell: bash
	env:
	RUN_IDS: ${{ inputs.run-ids }}
	run: Rscript ./R/delete_current_year_model_runs.R ${{ inputs.run-ids }}
	shell: bash

[jeancochrane]

This would be indeed be cleaner! I decided to switch the format of the run-ids input variable to expect a space-separated string in a4895c2, however, so I made a slightly different change here, instead using the RUN_IDS env var to strip spaces and replace them with commas. (My thinking is that spaces provide a clearer interface for the GitHub workflow, but on the off chance we ever need to run R/delete_current_year_model_runs.R manually, a single space-delimited string would be a much more confusing command line interface than a single comma-delimited string. Accepting a space-delimited string in the GitHub UI but a comma-delimited string via the command line seems like a reasonable compromise to me.)

[dfsnow]

I think it's fine! Though we might want to use a space-delimited string in order to match build-and-test-dbt. It would have to be quoted i.e. not positional args. I'm really fine with either though as long as it's clear what the input is supposed to be.

[dfsnow]

suggestion: IMO it'd be really helpful here to have a default input showing an example (nonexistent) run ID.

[jeancochrane]

True, done in a4895c2! Note that the trade-off here is that default arguments are populated as actual values in the workflow input form, rather than placeholder text, so there's a chance of a caller misinterpreting the form and submitting it with the (fake) default values. This should just raise an error and not actually delete any data, however, so the risk is worth it in my view.

[dfsnow]

That works for me! Let's also delete other test runs created during the model infra construction.

[dfsnow]

Totally! I would bump the cutoff month to May though, as sometimes we still do model touchup stuff in April.

[jeancochrane]

Thanks for the excellent review @dfsnow! I made some tweaks to the interface based on your comments, so I'm requesting one final review to make sure the new interface looks good to you. I tested via a pull_request event and confirmed that the new interface works as expected.

[dfsnow]

Looks good to me @jeancochrane!

[dfsnow]

praise: Good thinking!