Prometheus Scrape Probe¶

The prometheus_scrape probe pulls Prometheus /metrics endpoints — node_exporter, appliance exporters, any application exposing the text format — and ingests the samples through the agent pipeline, out to every configured output. It is the pull-side twin of the OTLP receiver: one agent per site collects both push (OTLP) and pull (Prometheus) sources.

Quick start¶

probes:
  - name: exporters
    type: prometheus_scrape
    params:
      targets:
        - "http://localhost:9100/metrics"      # node_exporter
        - "http://10.0.0.5:9117/metrics"       # appliance exporter
      interval: 60

Parameters¶

Parameter	Default	Description
`targets`	required	List of exposition URLs
`interval`	`60`	Seconds between scrape cycles
`timeout`	`10`	Per-target budget in seconds
`metric_match`	none	Regexp filter on metric family names; non-matching families are skipped
`bearer_token`	none	Sent as `Authorization: Bearer ...`; use `${env:...}` substitution
`insecure_skip_verify`	`false`	Accept self-signed exporter certificates

Targets are scraped in parallel (bounded). An unreachable exporter is a measurement (senhub.promscrape.up = 0), never a probe failure.

Behavior¶

Pass-through naming. Scraped metric names and labels are forwarded unchanged; counter/gauge semantics are preserved. Untyped samples are treated as gauges.
Scalars only. Counter, gauge and untyped series are ingested. Histogram and summary series are dropped and counted in senhub.promscrape.dropped — the same contract as the OTLP receiver.
Size cap. A scrape response is read up to 32 MiB; a runaway exporter cannot exhaust the agent's memory.

Self-metrics¶

One series per target (target tag).

Metric	Description
`senhub.promscrape.up`	1 when the target answered with a parseable exposition
`senhub.promscrape.scrape.duration`	Scrape wall-clock time
`senhub.promscrape.samples`	Scalar series ingested in the last scrape
`senhub.promscrape.dropped`	Histogram/summary series dropped in the last scrape

Operational notes¶

Use metric_match on big exporters. A full node_exporter exposes 1000+ series; metric_match: "^node_(cpu|memory|filesystem|network)" keeps cardinality under control.
Re-exporting to Prometheus. Scraped series flowing back out the agent's own Prometheus endpoint are prefixed senhub_, so they never collide with a direct scrape of the same exporter.
PRTG / Nagios outputs. These sinks key series per target; finer per-label splits are carried on the Prometheus and OTLP outputs.