Empty commits from failing token resets can pile up enormously

[h-vetinari]

I'm just noticing lots of failed token reset commits here. CC @beckermr @jaimergp

[jaimergp]

Travis is acting up:

failed to reset token for 'sweet-chem-o-mine': HTTPError('403 Client Error: Forbidden for url: https://api.travis-ci.com/auth/github')
[main d8bad2f] Keeping requests/sweet-chem-o-mine-token-reset.yml after failed token_reset
failed to reset token for 'qt-main': HTTPError('403 Client Error: Forbidden for url: https://api.travis-ci.com/auth/github')
[main 83a28b2] Keeping requests/qt-main.yml after failed token_reset
failed to reset token for 'python-ldap': HTTPError('403 Client Error: Forbidden for url: https://api.travis-ci.com/auth/github')
[main e8f4e44] Keeping requests/python-ldap-token-reset.yml after failed token_reset

[h-vetinari]

Where are you seeing the logs? I couldn't find at first glance where they're running. 😅

[h-vetinari]

In any case, in the spirit of conda-forge/conda-forge.github.io#1521 & conda-forge/conda-forge.github.io#2607, we should remove Travis here ASAP IMO

[jaimergp]

It could also be a missing auth token, although Travis should be able to obtain its own via Github.

Relevant code in smithy: https://github.com/conda-forge/conda-smithy/blob/5cb17421332e502c934c4cafddc288188604a794/conda_smithy/feedstock_tokens.py#L662-L709

[jaimergp]

We should also prevent the "failed run, keep file" operation from re-triggering the CI immediately. This has originated a cascade of errors since #2081, with one run every minute generating a commit every minute and so on.

[jaimergp]

Where are you seeing the logs? I couldn't find at first glance where they're running. 😅

Logs are in https://github.com/conda-forge/admin-requests/actions/workflows/main.yml

[h-vetinari]

This has originated a cascade of errors since #2081, with one run every minute generating a commit every minute and so on.

Yeah, I was wondering why I was 20k commits behind main here 🤣

[h-vetinari]

Logs are in https://github.com/conda-forge/admin-requests/actions/workflows/main.yml

Thanks; I had thought I must be looking in the wrong place because all those were passing 🤦

[h-vetinari]

We should also prevent the "failed run, keep file" operation from re-triggering the CI immediately.

Might be as easy as not allowing empty commits (or using some exponential cooldown)

[jaimergp]

~22070 commits as a result of these three weeks of unattended failures....

I assume we are not adding the [skip ci] bit because it needs to be retriggered in the next scheduled run. But yea, disallowing empty commits sounds good I think?

[jaimergp]

--allow-empty was added in 1bb6570 by @beckermr, as hinted in https://github.com/conda-forge/admin-requests/pull/648/files#r1069185785. I think we need to revisit that decision. This repo is now at 30K commits with most of those commit from unnecessary retry loops.

[h-vetinari]

To be fair, 22k out of those 33k are due to one(!) issue with travis that went undetected for a while.

[jaimergp]

It has happened before, pages and pages of repeated commits:

• https://github.com/conda-forge/admin-requests/commits/main/?after=9437fc2948884838bae911bc1afc3a63c3a1bb10+24175
• https://github.com/conda-forge/admin-requests/commits/main/?after=9437fc2948884838bae911bc1afc3a63c3a1bb10+25175
• https://github.com/conda-forge/admin-requests/commits/main/?after=9437fc2948884838bae911bc1afc3a63c3a1bb10+26000
• https://github.com/conda-forge/admin-requests/commits/main/?after=9437fc2948884838bae911bc1afc3a63c3a1bb10+30035