Skip to content

Hardware Classification Guide

Hardware classification automatically assigns profiles, workflows, and parameters to machines based on their hardware attributes. This eliminates manual profile assignment and ensures machines of the same type get consistent configuration.

See the Classification Operator Guide for the full rule syntax and operator-facing documentation.

How Classification Works

During discovery, DRP collects hardware inventory via gohai-inventory. Classification rules evaluate attributes from this inventory (manufacturer, model, disk count, memory size, NIC type) and apply matching profiles to the machine.

Classification Approaches

The simplest approach uses the universal pipeline's built-in classifier. It uses cl_process_params_to_profile to automatically construct and apply profiles based on combinations of universal/hardware, universal/application, rack/bom, and related params. Profile names follow a compound pattern:

  • universal-hw-<hardware>
  • universal-hw-<hardware>-<application>
  • universal-hw-<bom>-<hardware>-<application>

For example, if universal/hardware is PowerEdge_R750, the classifier looks for a profile named universal-hw-PowerEdge_R750 and applies it if it exists. More specific profiles (with bom and application segments) take precedence. No custom rules are required — just create profiles with the matching names.

Building a Classification Strategy

  1. Baseline representative machines of each type to understand the distinguishing hardware attributes
  2. Choose matching criteria — model name is usually sufficient; add disk count or memory if you have sub-configurations within the same model
  3. Create hardware profiles containing the target parameters (RAID, BIOS, firmware) for each machine type
  4. Define classification rules (or use universal-hw-* naming) to map attributes to profiles

Considerations

  • Classification runs early in discovery — profiles must exist before machines are classified
  • Test rules against gohai-inventory data from representative machines before deploying
  • Use the universal auto-classifier when possible; custom rules add maintenance overhead
  • Classification profiles can be packaged into content packs for multi-site deployment