Cut AWS testing costs by 75%

[unito-bot]

┆Issue is synchronized with this Jira Story
┆Issue Number: TOIL-1673

[unito-bot]

➤ Adam Novak commented:

We should reduce the number of CI runs or the fraction of CI runs that hit AWS for running clusters. We need to look at the AWS billing reports to make sure really we are spending all this on compute for the tests that use clusters, like we think we are.

To reduce CI runs we could not bump every PR when any other PR merges, or we could not run the same tests both before and after a merge, or we could not require a passing back-merged commit in the branch to merge.

[unito-bot]

➤ Adam Novak commented:

We could try and restrict dependabot: it has run tests 7 times to test an htcondor update that isn’t even under test.

[unito-bot]

➤ Adam Novak commented:

Maybe dependabot can be told not to update particular packages (google, htcondor, dev stuff like flake8, cluster-integration packages) that we know aren’t under CI.

[unito-bot]

➤ Adam Novak commented:

We could also make dependabot auto-merge bumps to dependencies that are actually tested instead of waiting for us to take it and making more test runs while we ignore it.

[unito-bot]

➤ Adam Novak commented:

https://docs.github.com/en/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file#ignore ( https://docs.github.com/en/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file#ignore|smart-link )

[unito-bot]

➤ Adam Novak commented:

To actually decide if this worked we might need to compare a before month and an after month, and control for time somehow.

[unito-bot]

➤ Adam Novak commented:

We think this probably has been successful.