Troubleshoot Kubernetes Inference Workloads with Datadog
Datadog, Tuesday, November 10th, 2026
Datadog previews Kubernetes inference workload monitoring that groups the resources behind a served model for troubleshooting.
Datadog has released Kubernetes inference workload monitoring in Preview, which connects the configuration declared in custom resources with the Kubernetes resources they manage.
In Kubernetes Explorer, an inference workload is a virtual resource grouping the pods, deployments, replica sets and custom resources that support a serving model.
It supports vLLM-backed workloads using KubeRay, NVIDIA Dynamo and KServe, and automatically tags detected pods so metrics, traces and logs can be correlated.
A Components tab shows readiness and pod counts per serving component, a Metrics tab shows runtime and resource usage, including GPU metrics for Dynamo workloads, and inference workloads are also exposed through Datadog's Kubernetes MCP tools.