pull model
A way of collecting metrics where the monitoring server fetches them from each service, instead of each service sending (pushing) them.
In more detail
Pulling makes a dead target visible, because the fetch fails; lets you reproduce any scrape by hand; and keeps load under the server's control. Its costs are that short-lived jobs may finish before they are scraped, and the server needs network access to every target.
Related terms
- scrape — One fetch by Prometheus of a service's metrics page over HTTP, whose values are then stored with a timestamp.
- Pushgateway — A Prometheus component that short-lived jobs push their metrics to, so Prometheus can scrape them after the job has exited.
- up metric — A series Prometheus writes for every target after each scrape: 1 if the fetch worked and 0 if it failed.
Taught in
- Prometheus Scraping And The Pull Model — Full-Stack Observability: Metrics, Tracing & Logging
- Off By Five: What Prometheus Is Not For — Prometheus Certified Associate (PCA) Exam Prep