Cloudy With a Chance of Kubernetes: Going From One To Three Cloud Providers
Talkat KubeCon + CloudNativeCon Europe 2025 · on Thursday, 2025-04-03· at 17:30· Level 0 | ICC Capital Hall | Room J·
summary
Datadog moves from one region on AWS to six regions and 29 edges across three cloud providers by running hundreds of clusters and tens of thousands of nodes on Kubernetes. They build a Software Delivery Platform that runs temporal workflows to roll updates region-by-region, replace Terraform-managed control planes with pods, and implement CRD-backed node groups plus Cluster Autoscaler. Result: most apps migrate to Kubernetes but provider mismatches (storage drivers, L7 load balancers, APIs) still force platform-level workarounds.
AI-written summary · about
chapters (22)
- 0:00Multi-cloud Kubernetes journey intro
- 1:24Reasons for three cloud providers
- 2:30Datadog infrastructure history
- 4:14Kubernetes for multi-cloud
- 6:02Managed K8s challenges 2018
- 7:29Custom K8s control planes
- 8:53Software delivery platform
- 10:18Cross-cluster infra management
- 11:59K8s control plane evolution
- 13:22Regional control plane deployment
- 14:03Abstraction limits and primitives
- 15:09Compute layer challenges
- 18:42Single pod per node limits
- 20:07Pod resource requests and limits
- 21:51Instance fleet sizing and costs
- 22:33Storage abstraction with CSI
- 24:57Networking and CNI evolution
- 27:44Cloud service API inconsistencies
- 29:51Database abstraction strategies
- 30:56Non-tech multi-cloud challenges
- 31:37K8s limitations and leakage
- 32:41Future perspectives and ongoing improvements
related talks
41mAn Adventure in Multi-Cloud with Kubernetes
35mKubernetes journey
33mKubernetes-Driven Multi-Cloud Deployments: Maximizing Scalability and Flexibility
41mKubernetes the Very Hard Way
27mKubernetes Everywhere: Lessons Learned From Going Multi-Cloud
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data
44mKubernetes at Datadog Scale
28m03 Kubernetes Evolution, from the Hard Way to Managed Multi-cloud