Problem
Failures were discovered by users first — no metrics, no centralized logs, no alerting on workload health.
Solution
Deployed workloads to Kubernetes with resource limits and probes, then layered Prometheus scraping, Loki log aggregation, Grafana dashboards per service and Alertmanager routes for on-call notification.
Architecture
- 01K8s deployments with liveness/readiness probes and limits
- 02Prometheus scrape + recording rules → Alertmanager routing
- 03Loki log pipeline + Grafana dashboards per service