LinkedIn Replaces Argo CD with a Custom GitOps Engine for Kubernetes Deployments
LinkedIn moved application deployments from per-cluster Argo CD instances to an internal engine that runs each deployment in an isolated pod, with no shared controller state.
The team describes watch contention, stale cache entries and field-management behavior among the problems it encountered at its scale. The replacement uses full-object PUT updates with conflict retries, a three-way drift check and per-deployment convergence monitoring. Argo CD remains in use for cluster add-ons and system services; the replacement is not described as open source.
LinkedIn reports fewer deployment-infrastructure failures and faster failed-deployment detection, but the year-apart comparison also includes improvements to the surrounding orchestrator, so the results cannot be attributed solely to the new engine.