A staging cluster that's been running for six months isn't really "staging" anymore. It has leftover deployments from features that shipped in March, config drift from a debugging session nobody rolled back, and permissions someone granted temporarily and forgot about. Nobody trusts it, so nobody wants to run their tests against it — which defeats the point of having it.

The alternative isn't "be more disciplined about cleanup." It's removing the thing that needs cleaning up in the first place. A cluster that's created for a single test run and torn down when that run finishes can't accumulate drift, because it never lives long enough to.

This is the model Rackmint is built around. Every cluster has a time-to-live by default — fifty minutes, unless you ask for something different — and gets deleted automatically when it expires. You're not relying on a CI job to remember to clean up after itself, and you're not paying for compute that's sitting idle between test runs.

The tradeoff is that this only works if creating a cluster is fast and cheap enough that "just make a new one" is genuinely easier than maintaining an old one. That's the actual engineering problem: not "how do we delete old clusters," but "how do we make a fresh one show up in seconds," so throwing the old one away stops feeling wasteful.

For workflows that do need a cluster to persist — a demo environment, an integration your team leans on daily — that's a different, deliberate choice, not the default. Disposable is what test infrastructure should default to; permanent is something you should have to ask for.