Mastering Efficiency in Argo CD: Scaling Smarter, Not Costlier
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 11:15· Platinum Suite | Level 3 | Room 1-2·
summary
A measured GCP experiment runs Argo CD on a 3‑node control plane managing 800 apps and costs ~$40/day (~$1.49/app, ~30% network). Key levers: host Argo CD on a dedicated HA control plane (three nodes), avoid inlining large Helm values and cap app history, reduce Redis replication chunk to 64MB and shard resource keys at ~50 to cut traffic ~10x, stop storing health metadata via controller.resourceHealthPersist, exclude Endpoint/EndpointSlice watches, and enable repo server concurrency with tuned parallel limits.
AI-written summary · about
chapters (8)
- 0:00Introduction and scaling
- 2:58Cost analysis experiment
- 7:16Understanding cost drivers
- 10:55Networking optimizations
- 16:32Reducing Redis and API traffic
- 21:06Optimizing compute and memory
- 23:32Repo server performance
- 27:11Efficient cluster management
related talks
18mUnveil the Secret Ingredients for Argo CD in the Enterprise-Scale
34mInfrastructure at Scale Using ArgoCD
10mKeynote: Pushing the Limits: Scaling ArgoCD’s Repo-Server To Handle 30...
15mA Quantitative Study on Argo Scalability
32mHow I Tricked ArgoCD Into Sharding on a Single Cluster
23mArgo at Scale: Navigating Complex Multi-Dimensional Deployments Across H...
29mControl Plane, Service, or Both? – Argo CD Multi-Cluster Architectures
09mLightning Talk: The $10,000 Argo CD Mistake: Eliminating Phantom Sync...