Engineering· 6 min read
Fifteen-minute deployments: what changed
A two-day release ritual became a fifteen-minute pipeline. The tooling was the easy part — here is what actually mattered.
NNAFATECHS AdminAdministrator
When we first met one of our clients, a release took two days, needed four people, and was scheduled around everyone’s nerves. Fifteen minutes and one button later, they deploy several times a week. The tools were ordinary. The change was in how the work was organised.
Make the pipeline the only path
If anyone can deploy by hand, the pipeline is advisory. We made the automated pipeline the only route to production, which forced every manual step to be either automated or deliberately removed.
Shrink the change
Large releases are risky because they contain many changes and any of them could be the culprit. Smaller changes, shipped more often, are individually boring. When something fails, the suspect list is short and rollback is trivial.
Test what actually breaks
A forty-minute test suite nobody trusts is worse than a four-minute one that everybody does. We pruned slow, flaky tests, kept a fast layer that runs on every commit, and moved the slow checks to a parallel stage.
Rehearse recovery
The confidence to release often comes from knowing you can undo it. We practised rollbacks until they were routine, added health checks that block a bad release automatically, and documented the runbook for the rest.
Measure the four numbers
Deployment frequency, lead time for changes, time to restore and change-failure rate. Watching them together stops teams optimising one at the expense of another, and gives leadership evidence rather than anecdotes.
