Managing open PRs/issues

[echoix]

I'm filing this for discussion at a PSC meeting, as a possibility for #3235.

I imagine that part of a solution could be with using a specially configured stale bot.
https://github.com/actions/stale

Even if one day, it can be used to close PRs/issues, it can be configured to not close and ping back after a time. So here are some specific configuration keys that might be interesting for us:

• start-date Only consider issues/PRs that are newer than that date. That would be my first iteration, so that we could grandfather all what exists, and apply the new behaviour from when we change our policy. After a while we could go back and enlarge the range, such that we don't overload ourselves.
• days-before-stale Time before the first reminder
• days-before-close: When set to -1, it will never close after the reminder.
• exempt-draft-pr When we enlarge the time span, we could use this option. However it consumes an extra API call, so we might want to avoid it when possible.
• exempt-milestones or exempt-pr-labels or exempt-issue-labels could be used in conjunction with my other discussion point, in Discussion about using a `backlog` milestone for older issues and PRs #3364

Remember that the message should include instructions on removing the stale label to have the timer restarted, and what label to add to exclude that issue/PR from getting stale. It's an escape strategy.

So, as for the smallest possible implementation steps that it could be split into, I imaging the following:

1. Only new PRs should be pinged back, but would never be actually closed after the timeout.
2. Same thing, for issues too
3. Expand the start date to include more issues. We might want to also dramatically lower the operations-per-run to limit the PRs that will be messaged, such as going through the PRs will take more than a week or so. I would choose something that will limit the number of messages to PRs like 2-5 per day.
4. Start actually closing issues/prs after the timeout. They can still be reopened after this, it's not fatal to have one closed.
5. Adjust the numbers of days to something a little bit stricter.
6. Remove the start-date exclusion, since we would be about finished with the backlog.

[echoix]

@veroandreo wou might want to take a look at it and add it to the discussion