Okay, so I've been deep in a multi-cluster setup for a new SaaS product, and the config drift issue is real. We're talking dev, staging, and production clusters, all needing slightly different configs for the same apps. I started with Flux but recently ran a parallel pilot with ArgoCD to compare.
My take? ArgoCD feels more intentional about multi-cluster from the ground up. Its "ApplicationSet" concept is a game-changer for managing drift. You define one template and then use generators to spawn apps per cluster, injecting cluster-specific values (like ingress hosts or resource limits) from a single source. If you need to update a config across all clusters, you change the generator or the template, and ArgoCD handles the sync per target. The UI also gives you an immediate, visual cluster-by-cluster health status, which is huge for my peace of mind.
Flux with Kustomize can *do* multi-cluster, sure, but it feels more like you're orchestrating it yourself. You're managing separate Kustomize overlays per environment and relying on your Git repo structure and Flux's Kustomization resources to point to the right place. It's flexible, but I've found it easier for drift to creep in if your overlay structure isn't perfectly disciplined. The reconciliation feels more decentralized.
Has anyone else hit a specific snag with drift in either tool? Like, maybe with secrets management across clusters or handling a cluster that's temporarily unreachable? I'm especially curious about how each tool reports and remediates that kind of state divergence.