Better Data from Workflow Runs?

[geoffreywiseman]

In theory, for a given repository you could walk back through workflow runs and for each run, get usage. That would allow:

• Different timespans than the billing period
• Rollup in different ways, like triggering event (is it pushes that are triggering the workflow usage?)

Might not make sense to do that across a whole organization -- could result in a LOT of API calls. But maybe something that would make sense for a single repo / or single workflow?

[estahn]

I'd like to filter out certain workflow runs by various filters, e.g. actor. In this case I would be able to see how much impact "dependabot" has on the action minutes. Given our test suite runs 8-10min atm, this is substantial if we do daily, weekly or monthly updates.

[geoffreywiseman]

I haven't yet done a proof-of-concept on this, to see if using the workflow runs as a data source would really work to get more details. Still looks plausible to me, including some actor information.

Maybe something like:
gh actions-usage --detailed --output=tsv codiform/gh-actions-usage

I might still need to make some decisions about how far back to go, whether or not to support going across multiple workflows / multiple repositories.

Sounds like there's a limit of 5k requests/hr, which is a fair amount, but also don't want to burn through a large percentage of that budget with a single CLI operation, so obviously need to make the right tradeoffs there. Looks like the listing of workflow runs would be several requests and then each workflow run would be another request -- not hard for that math to add up. Can imagine that a very active repo could have many runs in a month.