Prometheus
Prometheus is the CNCF metrics database and scraper ecosystem for time-series monitoring and alerting. We hold it for new work: custom metrics effort belongs in OpenTelemetry, and self-hosted Prom stacks rarely stay correct or used.
Blurb
An open-source monitoring system with a dimensional data model, flexible query language, efficient time series database and modern alerting approach.
Summary
What it is: Pull-based scraper, TSDB storage, PromQL, alert rules, and federation patterns for Kubernetes and services. Still fine as a storage/query target behind an OTel pipeline.
When to use (existing only):
| Situation | Notes |
|---|---|
| Inherited kube-prometheus-stack | Keep running; do not expand custom Prom client metrics |
| Prom-compatible remote store | Accept remote write from OpenTelemetry Collector |
When to skip (new work):
- Greenfield metrics: instrument with OpenTelemetry, not Prometheus client libraries as the primary model
- Multi-cloud / hybrid estates: prefer Datadog (trial) over a DIY Prom + Grafana stack
- Long-term log archival: Loki or Elasticsearch, not Prometheus
Trade-offs: “Just enough” scrape-and-PromQL is attractive early. Teams usually mis-model metrics and under-use what they collect. Time spent on custom metrics is better spent on OTel conventions and exporters.
Details
| Topic | Notes |
|---|---|
| Deploy | Single binary, Operator on Kubernetes, or managed compatible (Mimir, Cortex, AMP) |
| Pairing | Prefer OpenTelemetry for instrumentation; remote-write into Prometheus-compatible stores if needed |
| Successor path | New custom metrics → OTel; multi-cloud ops → Datadog; dashboards may still use Grafana |
References