Prometheus Is Metrics as a Query Language

Prometheus is a time series database whose interface is a query language. You scrape metrics. You store samples with labels. You ask PromQL questions.¹ Dashboards are a skin. I have used that query box as a user on other people’s clusters. I do not run Prometheus for mystic-bytes. Jekyll does not export a /metrics that would tell you whether an essay is true. Nightbind taught me the difference between a green graph and a question you can actually ask.

The pull model is the stance: the server scrapes you on an interval. Labels are the dimensions. PromQL lets you say rate, histogram quantile, this path, that status, without shipping a new panel first. That is why it won ops culture. Monitoring products sold charts. Prometheus sold a language for samples. Grafana sat on top and people forgot the language was the product.

Use it when you have numbers that must be sliced after the fact. Do not scrape a static site to feel like SRE. I keep Grafana out of the thesis on purpose. A chart you cannot rewrite as a query is decoration. I read the PromQL docs so I can talk to people who run scrapes. I still debug mystic-bytes with a build log.

— JV · Dark Heart Labs.

References

Prometheus Docs, “Querying basics,” https://prometheus.io/docs/prometheus/latest/querying/basics/. PromQL over labeled time series — the language, not the chart, as the interface. ↩