Building a GitOps Platform Exposed Hidden Dependency and Resource Problems

A team spent about two weeks and 194 commits building a GitOps platform that provisioned isolated customer evaluation environments directly from Git using ApplicationSets, Argo CD, and Helm. The provisioning flow worked, but operating it revealed problems such as migrations running too early and applications starting before their dependencies were reachable. Production incidents included a Redis/NATS startup race that left readiness checks returning HTTP 503 for more than three minutes across two client environments. The team responded by adding a TCP-connectivity gate across 23 Deployments, explicit deployment ordering, and changes to migration recovery. The authors conclude that reliable GitOps requires attention to dependencies, configuration compatibility, resource allocation, and recovery behavior, not just declaring desired state.