Projects05 Migration
Observability Migration Factory
A workflow that turns legacy monitoring definitions into reviewable open-source metrics, logs, rules, and dashboards.
Problem
Moving off a vendor monitoring stack is usually a manual rewrite of each check, rule, and dashboard, so intent gets lost and the cutover stays risky.
Solution
Built a migration workflow that ingests legacy monitoring definitions, maps that intent onto modern telemetry, generates configuration, validates the output, and opens a reviewable change before anything is applied.
Architecture
- Legacy config
- Parser
- Normalized model
- Generate
- Validate
- Change review
- Deploy
Technology
- Python
- Prometheus
- Grafana
- Loki
- Terraform
- GitHub Actions
Engineering decisions
- Mapping checks that do not share a single vendor model.
- Validating generated rules and dashboards before they page anyone.
- Idempotent generation so a second run does not create a different change.
- Keeping a person in the review step so a bad mapping cannot ship itself.
- A rollback path that restores the previous monitoring definition.
Automation
Discovery, mapping, and file generation run in CI. A person still approves the pull request.
Reliability
Invalid output fails the pipeline. Nothing is applied until the change is reviewed, and the previous definition can be restored.
Security
Vendor exports are sanitized before they are stored. Generated files do not carry hostnames or credentials from the source.
AI layer
A language model can suggest a mapping; a human still reviews every generated change. The factory runs without that step.