Autodiscovery of Prometheus endpoints for scraping

[binarylogic]

Feedback from our chat

• 90% of prometheus endpoints require autodiscovery (various orchestration systems like k8s, etc - most used cases) and it's not supported in vector.
• in most of cases it's better to use native collectors (or compatible ones like https://github.com/VictoriaMetrics/VictoriaMetrics/tree/master/app/vmagent) and receive benefits - delivery guarantees, deduplication etc. (especially considering that vector native proto does not have delivery guarantees so far)

I tend to agree. This would be a really nice feature for anyone using Prometheus, but it's not easy either :). For reference, here's the Prometheus configuration. Notice all of the cloud settings that enable autodiscovery.

We definitely need to discuss how we can reduce scope of this feature.

/cc @afoninsky

[MOZGIII]

Related: https://github.com/timberio/vector/blob/k8s-rfc/rfcs/2020-04-04-2221-kubernetes-integration.md#discover-and-gather-prometheus-metrics-for-kubernetes-api-resources

[zhongzc]

Would like to know if any progress on this feature request.

[jszwedko]

Unfortunately not, @zhongzc

[jszwedko]

(didn't mean to close this 😄 )

[waynec86]

FWIW, grafana-agent uses the library in the loki repo for scraping consulagent targets for services running on the local host here: https://github.com/grafana/loki/blob/main/clients/pkg/promtail/discovery/consulagent/consul.go

Would be great if vector had this so we can have one agent that can export all our metrics/logs/traces.