What does rate(windows_net_bytes_received_total[1m]) compute, and what does the [1m] mean?

[1m] selects every sample of the counter from the last minute (a range vector). rate() turns that into the average per-second increase over the window. Here: received bytes per second, so network throughput.

* The window grabs the last minute of samples; rate() turns their rise into bytes per second. *

Step by step:

1. windows_net_bytes_received_total is a counter: total bytes received since the exporter started, one series per network interface.
2. [1m] makes it a range vector: for each series, all samples from the last minute instead of only the newest one.
3. rate() takes the increase across that window and divides it by the window length in seconds, giving a per-second rate. It also handles counter resets: if the value drops (exporter restart), it treats that as a reset rather than negative traffic.

The window trades responsiveness for smoothness. [1m] reacts quickly but is spiky; [5m] is smoother. It must span at least two scrapes; with the default 15-second scrape interval, one minute holds about four samples.

Grafana's Explore view suggests rate() automatically when you pick a counter such as node_cpu_seconds_total, for exactly this reason.

Go deeper:

• Prometheus — Query functions: rate() — the definition, with the advice to use it on counters only.
• PromLabs — How exactly does PromQL calculate rates? — extrapolation and counter resets explained with diagrams.