The SRE Guide to Kubernetes Observability: Red vs. Use Methods
The article compares the RED (Rate, Errors, Duration) and USE (Utilization, Saturation, Errors) methods for Kubernetes observability from an SRE perspective. RED focuses on user-facing request metrics, while USE targets resource health, helping teams choose the right approach for monitoring microservices and infrastructure.
Background
- RED (Rate, Errors, Duration) and USE (Utilization, Saturation, Errors) are two complementary frameworks for monitoring systems, originally developed by Google SREs and Brendan Gregg respectively.
- RED focuses on user-facing service health (request-level metrics common in microservices), while USE focuses on resource health (CPU, memory, disk, network) at the infrastructure level.
- SRE (Site Reliability Engineering) is a discipline from Google that applies software engineering to operations, ensuring systems are reliable, scalable, and observable.
- Kubernetes (K8s) is the dominant open-source container orchestration platform; its dynamic, ephemeral nature (pods spinning up/down, services scaling) makes traditional monitoring approaches harder to apply.
- The article argues that neither RED nor USE alone is sufficient for Kubernetes observability — combining both is needed: USE for the underlying node/container resources, RED for the application services running on top.