Skip to content

//DevOps

The no-deploys-on-Friday rule is a symptom

6 min read

We have walked into a lot of teams with a release ritual. A window gets announced, a few people stay late, and nobody ships on a Friday. The ritual is not the problem. It is a reasonable response to a real one.

What the rule is really saying

If you are afraid to deploy on a Friday, you are not afraid of Fridays. You are saying that you do not know whether the build is good, and that if it is not, getting back to safety will take longer than the weekend.

Both halves matter, and most teams try to fix the first one first. They add tests, they add review gates, they add a checklist. That helps, but it treats every release as something to be survived. The faster win is usually the second half.

Rollback is a button, not a runbook

On one platform we worked on, the change that altered team behaviour most was not the test suite or the pipeline. It was making rollback a single click back to the last good image.

That sounds like a small feature. It changes everything about how people behave. When going back is cheap, going forward stops being frightening. The Friday rule on that team died within a couple of weeks, and nobody had to argue about it. It just stopped making sense.

A runbook is not the same thing. A runbook is a rollback you have described. A button is a rollback you have tested, repeatedly, in the same way every time.

The order matters more than the tooling

Get the build reproducible first. Same commit, same dependencies, same artefact, every time. Lock your dependencies and produce an immutable image rather than building on the target machine.

Then make the tests trustworthy. A flaky suite is worse than no suite, because it teaches people to re-run until green, which is the same as not testing. Fast feedback is the goal: lint and unit tests in seconds, integration checks in minutes, and a clear signal when something is wrong.

Only then automate the deploy. Teams that automate deployment before their tests are reliable have not built a safety net. They have built a faster way to ship bugs.

The step everyone skips

Provisioning and configuring environments from code is the piece that gets dropped when a deadline appears, and skipping it is exactly why “it works in staging” stops meaning anything.

Environments configured by hand drift. Somebody sets an environment variable during an incident and never writes it down. A package gets upgraded on one box. The drift is invisible until it is not, and then you are debugging a difference between two machines that are supposed to be identical.

Where this ends up

Every commit built, tested and containerised. Environments reproducible from code. Zero-downtime releases with a rollback that takes one click. Release cycles stop being scheduled events and become something nobody thinks about much.

That last part is the actual goal. The best sign that delivery is working is that people stop discussing it.