Skip to content

agent_api.md

Latest commit

 

History

History
76 lines (64 loc) · 2.88 KB

agent_api.md

File metadata and controls

76 lines (64 loc) · 2.88 KB

Endpoints exposed by the Agent

The core Agent exposes a variety of HTTP and GRPC endpoints that can be organized into two groups:

  1. Control: These endpoints are used to send commands that can control and inspect the state of the running Agent.
  2. Inter-process communication: These endpoints are used by the different agent processes to communicate between each other.
  3. Telemetry: These expose some internal telemetry that is useful for profiling and debugging.

Control API

This API is accessible via HTTPS only and listens by default on the localhost interface on port 5001. The listening interface and port can be configured using the cmd_host and cmd_port config options.

It is considered internal, endpoints can change at every version, and it should not be relied on by users.

Inter-process Communication API

Similarly to the control API, this API is accessible via HTTPS only, and listens by default on the localhost interface on port 5009. The listening interface and port can be configured using the agent_ipc.host and agent_ipc.port config options.

It is considered internal, endpoints can change at every version, and it should not be relied on by users.

Telemetry API

There are 3 different systems exposing data on the same port but at different endpoints. The default port is 5000 and can be configured by changing expvar_port.

ExpVar

Expvar is at /debug/vars

$ curl -s http://localhost:5000/debug/vars | jq '.scheduler'
{
  "ChecksEntered": 8,
  "Queues": [
    {
      "Buckets": 900,
      "Interval": 900,
      "Size": 1
    },
    {
      "Buckets": 15,
      "Interval": 15,
      "Size": 7
    }
  ],
  "QueuesCount": 2
}

Prometheus-style telemetry

Prometheus style telemetry is exposed at /telemetry if the config option telemetry.enabled is set to true.

$ curl -s http://localhost:5000/telemetry | head
# HELP aggregator__processed Amount of metrics/services_checks/events processed by the aggregator
# TYPE aggregator__processed counter
aggregator__processed{data_type="dogstatsd_metrics"} 1
aggregator__processed{data_type="events"} 1
aggregator__processed{data_type="metrics"} 102
aggregator__processed{data_type="service_checks"} 6
# HELP aggregator_tags_store__hits_total number of times cache already contained the tags
# TYPE aggregator_tags_store__hits_total counter
aggregator_tags_store__hits_total{cache_instance_name="aggregator"} 171
aggregator_tags_store__hits_total{cache_instance_name="timesampler #0"} 0

Pprof

Pprof is available at /debug/pprof. This endpoint has an index that lists the different pprof endpoints and the official go pprof docs can also be referenced.

$ curl -s http://localhost:5000/debug/pprof/profile?seconds=60 > ./cpu.out

Not documented here

  • non-core agent endpoints (i.e., what do the security, process, and cluster agents expose)
  • GRPC endpoints