Observability Practical By Samson Tanimawo, PhD Published Jul 17, 2026 4 min read

The OTel Collector Deployment Pattern That Scales

Sidecar, daemonset, or gateway? The deployment topology that handles 10M+ spans per minute without falling over.

Sidecar: per-pod collector

One collector per app pod. Cheap to debug; failure isolated.

Cost is high at scale. Each pod runs a collector even if it produces little telemetry.

One collector per node, shared across pods. Reduces collector count by 10-100x.

Failure affects all pods on the node; debug is harder. Standard pattern for production.

Final aggregation before vendor send. Handles routing, sampling, transformation.

Required at scale; the place where tail sampling happens.