The engineering decision
Begin with one bounded workload. Prove that the team can propose, approve, reconcile, observe and recover a change before extending the operating model.
On this page
Keep the desired state reviewable
The OpenGitOps principles describe declarative desired state that is versioned, automatically pulled and continuously reconciled. Git provides a useful review history, but the operating model also needs a controller, permissions and feedback from the running system.
GitOps does not make configuration drift or failed deployments impossible. The team still owns the intended state, the reconciliation behaviour and the response when reality differs.
Follow one change into the system
- Propose. An engineer changes one deployment setting in versioned configuration and states the expected result.
- Review. The configured checks run; an appropriate person reviews the change and its operational impact.
- Reconcile. The configured controller retrieves the approved desired state and attempts to apply it.
- Observe. The operating team sees reconciliation status and application health, then accepts, corrects or recovers the change.
For Kubernetes, GitLab documents its integration with Flux. Use the current setup guide for your repository or OCI source, controller and agent configuration. The sequence above is an operating example, not a deployable configuration.
Agree the boundaries before automating
- What is in scope: choose the workload and environments first.
- Who may approve: keep review and production access responsibilities explicit.
- How secrets are handled: keep credentials out of ordinary configuration and define rotation ownership.
- What failure looks like: distinguish a reconciliation error from an unhealthy application.
- How to recover: test a known-good desired state and any data or dependency constraints.
A Git revert may restore configuration intent; it does not necessarily reverse a database migration, external side effect or incompatible application change. Include those limits in the recovery plan.
Start small enough to learn
Use one non-critical workload and a named operating team. Exercise a normal change, a rejected change, a controller failure and a recovery. Decide what evidence and alerts the team needs before introducing more repositories or production responsibilities.
Expand only when the owners can explain the path from approval to running state. Record exceptions and update the operating notes as the model changes.
Make the handover part of the engineering
Useful delivery work includes repository structure, access boundaries, review checks, controller integration, observation and a rehearsed operating walkthrough. The acceptance criterion is that the customer team can operate the workflow after handover.
OTTRA can assess the starting point through a Health Check or scope a defined engineering project directly. Agree the workload, owners, outputs, timing and cost before work starts.
Primary sources
Source scope and product terms can change. Checked on 7 November 2026.





