Our entire test suite started silently failing Docker builds. No alerts. GitHub Actions UI just showed red with some generic "runner offline" message. By the time we SSH'd in, /var was at 99% because we weren't pruning old images.

Would've caught it instantly on GitHub Actions. Their runners just... work. You get disk cleanup out of the box. Cost like 3x more but honestly worth the peace of mind for a 15-person team.

The self-hosted thing only makes sense if you've got: dedicated devops person, custom hardware needs, or you're burning insane GitHub Actions minutes. We had none of those. We just wanted to feel fancy.

Next week we're killing the runners, moving back to Actions. Adding proper monitoring to the transition stuff we do need self-hosted for (signing, artifact caching). The ops tax on runners isn't worth it.