LEARN. PRACTICE. GROW.Concept and mechanism
GitOps adds a versioned source and an agent attempting to reconcile observed state. A manual change can disappear if the source still declares another configuration. During an incident, coordinate mitigation with that mechanism: update the approved source or use an authorized temporary suspension with an owner and defined resumption. Replica count is not the only success criterion. Observe queue depth, latency, errors, and downstream capacity. If six workers saturate the database, scaling merely moved the problem. Record the exception’s reason, accepted final configuration, and criteria for returning to normal capacity after the peak.
Guided application
Probes answer different questions. Readiness evaluates whether the Pod should receive normal traffic; readiness failure is not itself the restart action associated with liveness. A Running application may remain unready. For a restarted container, previous-instance logs can help when still available; select namespace and container and use --previous. Collect evidence before destroying objects. Deployment rollback also has boundaries: restoring a template does not undo an incompatible schema migration. Before reverting an image, assess data and dependencies with their owners. Recovery is complete when the transaction and relevant controls are validated, not merely when a command finishes.
A manual change from three to six replicas returns to three: first compare reconciled source and events before repeating the command.
Common pitfalls
Fighting the reconciler; forgetting resumption; Running confused with Ready; rollout undo treated as database recovery; deleting evidence early.
Related topics: Observability, objectives, and ecosystem · Desired state, control, and containers
A release needs coherent intent, functional evidence, and data-compatible recovery.
Reference: OpenGitOps principles · KCNA current four-domain curriculum; edition date unconfirmed