Engineering / devops
DevOps and delivery
A delivery pipeline is a control system: it should make the safe path easy and leave evidence when it says no.
Stages are decisions
Build, test, scan, release, and deploy are not boxes on a diagram. Each stage should answer a question that someone can audit later: what changed, what was tested, what was promoted, and by whom?
- Pin dependencies and identify artifacts by immutable digest.
- Keep environment promotion separate from compilation so a rebuild cannot silently change the thing being promoted.
- Make rollback a tested operation, not a sentence in a runbook.
Evidence over green ticks
A green pipeline is useful only when the checks behind it match the risk. For regulated systems that includes traceable approvals, dependency provenance, and retention that outlives the next deployment.
The pipeline should be able to explain why a release was allowed, not only that it was allowed.
Feedback loops
Fast feedback is not the same as skipping checks. It means running cheap, high-signal checks early and reserving slower integration and security checks for the artifact that will actually ship.