The HTTP API is designed for cases where you want to get metrics from one particular node or task, for example to build third-party tooling. It is exposed via adminrouter for each agent:
http://<your-dcos-master-url>/system/v1/agent/<agent-id>/metrics/v0
The /node endpoint shows metrics about a node. These are equivalent to the metrics available from PSUtil, and contain information about memory, cpu, and disk utilization.
The /containers endpoint yields a list of all container IDs running on the node. A container is the lowest common denominator for Mesos workloads. In general, one DC/OS task has one container (however some tasks can have multiple container IDs, and sometimes a container exists with no corresponding task).
The /containers/container-id endpoint gives a list of metrics related to a container’s resources. It is populated with data from the mesos /monitor/statistics endpoint.
Note that no data (HTTP 204) is returned for nested containers (for example tasks in a pod, or tasks created by a framework which is running on Marathon). This is because they are missing from the mesos statistics endpoint.
The /containers/container-id/app gives a list of metrics emitted by a workload inside a container over statsd, as described in Instrumenting your Code.
Note that no data (HTTP 204) is returned for containers for which no metrics exist (for example containers which use the Docker executor).