Kubernetes at Datadog Scale
summary
Datadog runs dozens of Kubernetes clusters, several with +1K nodes per cluster, managing its own control plane for deep debugging. Networking uses Cilium (eBPF) CNI to get direct host‑routing across multiple clouds, replace kube‑proxy and add host‑to‑host encryption — Datadog maintains contributions upstream. Developer-facing platform uses self‑service node group CRDs, a move toward shared “nodeless” pools, repacking via Cluster Autoscaler, and policy enforced with Gatekeeper (OPA). A proof‑of‑concept replaces VPA’s recommender with a custom VPA recommender wired to Datadog metrics.
AI-written summary · about
chapters (12)
- 0:00Introduction and overview
- 3:56Kubernetes growth and trends
- 7:32Why Datadog chose Kubernetes
- 9:00Pod networking and CNI plugins
- 13:40Adopting Cilium for multi-cloud
- 15:51Developer experience and resource model
- 20:10Self-service node provisioning
- 24:42Vertical pod autoscaling
- 25:45Transitioning to nodeless clusters
- 26:30Policy enforcement with Gatekeeper
- 33:19Building platforms on Kubernetes
- 34:46Q&A session
related talks
41mKubernetes the Very Hard Way
14mMonitoring Network Traffic in Kubernetes
34mCloudy With a Chance of Kubernetes: Going From One To Three Cloud Providers
26mScaling Network Policy Enforcement Beyond the Cluster Boundary with Cilium
39mReliable Data Replication from Postgres on Kubernetes
26mMonitoring Kubernetes: Follow the Data [B]
34mHow to Survive to High Demanding Services
01h 41mDatadog DASH 2019 Keynote