Skip to main content
On this page

Unrecommended Resource Attribute

Low Signal: Metrics

Problem

Your metrics carry resource attributes that add cardinality without adding observability value. Insights specifically checks for:

  • host.ip: changes on reschedule and multiplies series, while telling you little a hostname doesn’t.
  • host.mac: hardware addresses have no observability use.
  • os.description: long free-text strings that vary per image build.

These usually arrive via a default resource detector that captures everything it can see.

Impact

  • Each attribute value multiplies metric series and storage.
  • Backends index and store bytes nobody will ever query.
  • Small per-service waste, but it applies fleet-wide by default.

Example

Carrying dead weight:

# resource attributes
service.name: checkout
host.ip: 10.42.17.3
host.mac: 02:42:ac:11:00:02
os.description: "Linux 6.8.0-45-generic #45-Ubuntu SMP ..."

Trimmed:

# resource attributes
service.name: checkout
host.name: node-17

How to fix

  1. Configure the resource detectors to include only attributes you use; most SDKs let you pick detectors or deny-list attributes.
  2. Or strip them centrally with a Collector resource processor, which fixes the whole fleet in one place.

Expected gains

Slightly leaner metrics everywhere, and one less source of surprise cardinality: a small win that applies to every service at once.

See the Insights catalog for everything Insights can detect.