Challenge
The service was outgrowing its infrastructure: manual releases, a single application server with no redundancy and no unified monitoring. Before onboarding new partners it had to prove resilience and a controlled change process.
What we did
- Designed a network architecture with separate environments and a dedicated segment for payment components
- Deployed a Kubernetes cluster in Yandex Cloud across three availability zones
- Set up PostgreSQL redundancy and tested recovery from node and zone failures
- Built GitLab CI/CD: build, tests, zero-downtime release and one-step rollback
- Set up Prometheus and Grafana monitoring and alerting and prepared operating documentation
Outcome
A new version goes live in 15 minutes; it used to take half a day. Availability over 12 months is 99.95%. The failover test results were shared with the partner.