Why does switching from the Node Exporter to Telegraf in a running setup break alerts and dashboards?

The two collectors name their metrics (and labels) differently, and every alert rule and every Grafana panel is a PromQL query that selects metrics by those names.

An alert like node_filesystem_avail_bytes < … or a panel graphing node_cpu_seconds_total simply finds no data once the Node Exporter is gone, because Telegraf calls the same measurement something else. Every affected query has to be rewritten.

Lessons:

• Decide "Telegraf or Node Exporter?" at the start, before alerts and dashboards pile up.
• Grafana's dashboard library has dashboards made for the Prometheus + Telegraf combination, but a home-built setup usually needs manual adjusting after a switch.

Go deeper:

• Prometheus — Metric and label naming — why names and labels are the contract every query depends on.