Prometheus service discovery

nl6 publishes the running fleet as a Prometheus HTTP service discovery endpoint. Prometheus picks up devices as they are created and drops them when they are deleted or stopped. There is no static target file to regenerate.

The endpoint is always on. It needs no flag.

GET /api/v1/prometheus/sd

What the targets are

nl6 devices serve no /metrics endpoint. Each target is a device's SNMP agent, meant for snmp_exporter. Prometheus scrapes snmp_exporter and passes the device address as the target parameter.

The body is a JSON array with one target group per running device, sorted by IP:

[
  {
    "targets": ["10.42.0.1"],
    "labels": {
      "__meta_nl6_resource": "cisco_ios",
      "__meta_nl6_device_type": "Cisco IOS",
      "__meta_nl6_sys_name": "core-rtr-01",
      "__meta_nl6_snmp_port": "161"
    }
  }
]

Field	Meaning
targets	The device IP. On a port other than 161 it is ip:port. snmp_exporter accepts both forms.
__meta_nl6_resource	The device's resource file, lower-cased and without .json (for example cisco_ios). Add .json to use it in POST /api/v1/devices. Devices created with no resource file, such as the -auto-start-ip batch, report the startup default profile they serve (asr9k). Use it to pick an snmp_exporter module.
__meta_nl6_device_type	The human-readable device type label.
__meta_nl6_sys_name	The device's sysName. No two running devices in one simulator share a name, so you can use it as a device key. See Name uniqueness.
__meta_nl6_snmp_port	The SNMP UDP port.

An empty fleet returns []. Stopped devices are not listed.

Prometheus drops every __meta_* label after relabeling. Keep the ones you want by copying them to a plain label, as the example below does.

Name uniqueness

Each device draws a random sysName at creation. A draw that matches a name held by a running device is drawn again. If 64 draws in a row are taken, the device gets the first draw with its IP appended, for example loki-van-cisco-crs-x-172-27-0-1. A deleted device frees its name for later devices. Names are not stable across restarts. Uniqueness holds within one simulator process, not across several.

What the body does not carry

No credentials. The SNMP community and SNMPv3 passwords are never in the body. Prometheus logs and caches SD responses. Configure credentials in snmp_exporter's snmp.yml and select them with the auth parameter.

No module choice. nl6 does not guess which snmp_exporter module fits a device type. Map __meta_nl6_resource to __param_module in your relabel rules.

Prometheus configuration

This job discovers the fleet from nl6 on nl6:8080 and scrapes it through snmp_exporter on snmp-exporter:9116.

scrape_configs:
  - job_name: nl6-snmp
    http_sd_configs:
      - url: http://nl6:8080/api/v1/prometheus/sd
        refresh_interval: 60s
    metrics_path: /snmp
    params:
      module: [if_mib]
      auth: [public_v2]
    relabel_configs:
      # Hand the device address to snmp_exporter.
      - source_labels: [__address__]
        target_label: __param_target
      - source_labels: [__param_target]
        target_label: instance
      # Keep the nl6 labels you want on every series.
      - source_labels: [__meta_nl6_sys_name]
        target_label: sys_name
      - source_labels: [__meta_nl6_resource]
        target_label: nl6_resource
      # Optional: a different module for one device family.
      - source_labels: [__meta_nl6_resource]
        regex: juniper_.*
        target_label: __param_module
        replacement: if_mib
      # Scrape snmp_exporter, not the device.
      - target_label: __address__
        replacement: snmp-exporter:9116

To scrape a subset of the fleet, add a keep rule on any __meta_nl6_* label.

New and deleted devices show up on the next refresh_interval.

Reachability

snmp_exporter sends SNMP to the device IPs, so it needs a route into the simulated range (10.42.0.0/16 by default). GET /api/v1/devices/routes generates a route script for a Debian or Ubuntu host. See the web API reference.