Hosts

How hosts appear #

A host appears the first time your telemetry names it, in any signal. The host name is the identity, so a consistent host.name resource attribute is what keeps one machine from showing up as several.

• Spans, logs and metrics all contribute to the host list.
• A host that has stopped reporting leaves the list as its window passes.

Host detail #

• Environment and service — where the host is running and what it runs.
• Operating system — from your os.type attribute.
• Cloud provider and region — from your cloud attributes.
• Cluster — the Kubernetes cluster the host belongs to, when applicable.
• Instance type and agent version — where reported.
• CPU, load and memory — averaged over the selected window.
• Network traffic — bytes in and out over the window.
• Uptime — since the host was first seen in the window.

Health #

Health summarises the window. A host reporting an error signal — a failed span, an error-level log — is marked warning for that period; a host with no errors is ok.

Containers on a host #

A host's detail lists the containers reporting from it, with their CPU and memory. From there, open any container for its own detail page.

The host map #

The map places every host that reports metrics on a canvas, sized by load. Use it to spot the machine that is struggling rather than reading a table of them.

Onward #

A host that is struggling usually explains something about what runs on it: Containers for the workloads it hosts, Processes for the individual programs. To see the tile view, Host map.

Time ranges #

Host views accept a wide range — up to 400 days — so you can compare a quiet week against the same week under load.