Agent

Configure CostGraph agent

Similar to the CostGraph operator, the CostGraph agent supports additional configuration options for the agent deployment via conf.yaml.

Quickstart

prometheus:
  enabled:       true
  metrics_port:  9101
  url:           "http://localhost:9090"
  timeout:       10s 
  interval:      15s
  tenantHeader:  ""
  tenantValue:   ""
  bearerToken:   ""
  tlsInsecure:   false
  labels:
    job: costgraph-agent
expected_utilisation:
  cpu: 80
  memory: 80
api_key: ""
metrics_only: false

Assuming you have prometheus running within the same instance as the agent, you can use the configuration above.

Config reference

prometheus · object · required

Prometheus client configuration for the agent.

Show Prometheus Properties

enabled · boolean

Whether the Prometheus client endpoint is exposed by the agent.

Example:

metrics_port · number

Port on which the agent exposes Prometheus metrics.

Example:

url · string

Base URL of your Prometheus server that the agent queries for metrics.

Example:

timeout · string

Request timeout when querying Prometheus, in duration format.

Example:

interval · string

How often the agent queries Prometheus, in duration format.

Example:

tenantHeader · string

Optional HTTP header name used for multi-tenant Prometheus setups (e.g., Cortex/Mimir). Leave empty if not required.

Example:

tenantValue · string

Optional value for the configured tenant header. Leave empty if not required.

Example:

bearerToken · string

Optional HTTP Bearer token used to authenticate to Prometheus.

Example:

tlsInsecure · boolean

Skip TLS certificate verification when connecting to Prometheus. Example:

labels · object

Arbitrary key/value labels attached to the agent’s exported metrics and identity. Helpful for identifying instances when running a fleet of agents (e.g., per cluster, region, or host).

Show Labels Example

labels:
  job: costgraph-agent
  cluster: prod-eu-1
  node: ip-10-0-1-23

expected_utilisation · object

Expected utilisation targets used to guide right-sizing recommendations.

Show Expected Utilisation Properties

cpu · number

Target CPU utilisation percentage.

Example:

memory · number

Target memory utilisation percentage.

Example:

api_key · string · required

API key used by the agent to authenticate

Example:

metrics_only · boolean

When enabled, the agent exports metrics only and skips recommendation. Useful if you only need observability or are running the agent in a metrics-forwarding mode.

Example: