Platform Engineering · 2024 — Present
CI/CD Delivery Platform for 75+ Microservices
Replaced a manual SVN release process with a Jenkins and GitLab pipeline suite covering build, merge evaluation, deployment, backup and sub-minute revert.
- 60m → 10m
- Deploy downtime
- 60m → <30s
- Revert time
- 75+
- Microservices
- Weekly → on-demand
- Release cadence
The situation
TCS Enterprise Manager is a large ITSM/ITOM platform: 75+ backend microservices and five-plus frontends written against different frameworks. When I joined the team, all of it lived in SVN and shipped by hand.
Deployments happened every Thursday. There was no code review gate, no automated evaluation of a merge, no reliable backup step, and no revert path short of redeploying the previous artefact manually. A bad release cost roughly an hour of downtime, and undoing one cost about the same.
What we built
Another engineer and I were asked to move the unit to a CI/CD model.
Branching and delivery model. Before building any pipeline, we documented what branches mean, what a merge has to pass, how an environment is promoted, and who approves what. The pipelines were implementations of that document.
Merge evaluation pipelines. A merge request now has to pass build, static analysis and security scanning before it is eligible. Auto-merge pipelines handle the mechanical merges.
Deployment pipelines. Parameterised per microservice, with the deploy sequence, health checks and rollback trigger defined in the pipeline rather than in someone’s memory. We managed and cleaned the working directories ourselves instead of using Jenkins’ default workspaces.
Backup and revert pipelines. A revert went from about an hour of manual work to a single pipeline run finishing in under thirty seconds. After that the team stopped batching changes into the weekly window and deployed when a change was ready.
Governance as code
Each microservice carries a YAML descriptor next to its source declaring who can deploy it, which environments are valid targets, and which promotion paths are allowed. Server and infrastructure configuration is managed the same way, so access changes go through merge requests.
Result
Deployment downtime fell from about an hour to about ten minutes. Reverts fell from about an hour to under thirty seconds. The fixed weekly release window is gone, and other Platform Solutions teams have since adapted the same model.