Prometheus: Up & Running

Chapter 7. Node Exporter

The Node exporter¹ is likely one of the first exporters you will use, as already seen in Chapter 2. It exposes machine-level metrics, largely from your operating system’s kernel, such as CPU, memory, disk space, disk I/O, network bandwidth, and motherboard temperature. The Node exporter is used with Unix systems; Windows users should use the wmi_exporter instead.

The Node exporter is intended only to monitor the machine itself, not individual processes or services on it. Other monitoring systems often have what I like to call an uberagent; that is, a single process that monitors everything on the machine. In the Prometheus architecture each of your services will expose its own metrics, using an exporter if needed, which is then directly scraped by Prometheus. This avoids you ending up with uberagent as either an operational or performance bottleneck, and enables you to think in terms more of dynamic services rather than machines.

The guidelines to use when you are creating metrics with direct instrumentation, such as those discussed in “What Should I Name My Metrics?”, are relatively black and white. This is not the case with exporters, where by definition the data is coming from a source not designed with the Prometheus guidelines in mind. Depending on the volume and quality of metrics, tradeoffs have to be made by the exporter developers between engineering effort and getting perfect metrics.

In the case of Linux, there are thousands of metrics ...