Set up infrastructure

The attributes that make hosts, containers, pods, processes and functions appear.

1 min read Updated Oct 21, 2026 Sentinel Platform Team

infrastructuresetupattributestagsonboarding

Everything is named by your telemetry #

The infrastructure views describe what your telemetry describes. Nothing is discovered separately, so the views are only as complete as the attributes your collectors set.

The rule is that each entity has one attribute that identifies it, and every signal that sets it contributes to that entity.

The identifying attribute per entity

Host        host.name
Container   container.name
Pod         k8s.pod.name
Process     process.pid
Function    faas.name

A container is additionally qualified by its host, so two containers with the same name on different machines are two different containers.

What each view needs #

Hosts — host.name, plus os.type, host.type and the cloud attributes for the detail panel.
Containers — container.name, plus container.image.name for the image.
Kubernetes — k8s.pod.name, with k8s.pod.uid, k8s.namespace.name, k8s.node.name and k8s.cluster.name for full detail.
Processes — process.pid, plus process.command_line and process.owner.
Functions — faas.name, plus faas.coldstart, faas.max_memory and faas.trigger.

Any signal counts

Spans, logs and metrics all contribute. A container that only writes logs is still listed, which is why a service with no metrics is visible here at all.

Attributes that fill in the detail #

Optional, and what each one fills

cloud.provider          cloud.region          cloud.availability_zone
cloud.instance.type     cloud.account.id
os.type                 os.version
host.type               host.cpu.cores        host.memory.total
host.ip                 host.kernel.version
container.id            container.image.name   container.image.tag
k8s.pod.uid             k8s.namespace.name    k8s.node.name
k8s.cluster.name
process.command         process.command_line   process.executable.name
process.owner           process.runtime.name
faas.instance           faas.max_memory        faas.version
sentinel.team           the owning team, shown on the resource catalogue

sentinel.team is what makes the owner column on the resource catalogue populate. It is the one attribute where every reporter must agree.

Resources, not just metrics #

To see how much CPU, memory and load a host is using, send the corresponding metrics. The identity attributes make a host *appear*; the metrics make its numbers *mean* something.

Host metrics

system.cpu.user         CPU percentage
system.mem.used         memory in use
system.disk.in_use      disk in use
system.load.5           five-minute load

Container CPU and memory come from container.cpu.usage and container.memory.usage.total, with container.memory.limit as the denominator for the percentage.

Check your work #

Infrastructure → Hosts lists the machines your telemetry named.
Host map shows them grouped, which is the quickest way to see whether any arrived without the attributes you expected.
Resource catalogue lists every entity, including the ones with no metrics.

A host that disappeared

The views cover the window you are looking at. A host that stopped reporting leaves the window as that period passes, so widen the range before concluding a host is gone.

Then #

Once entities are appearing: Hosts for the list and its detail, Containers for what runs on them, and Kubernetes for pod state.