feat: Add hardware categorization design

docs/design-guide/device-types.md

@@ -0,0 +1,280 @@
+# Device Types
+
+Device types provide a structured way to define and categorize hardware models
+supported by UnderStack. They serve as a declarative specification of hardware
+characteristics, enabling consistent hardware identification, resource class
+mapping, and infrastructure automation across the platform.
+
+## Purpose and Architecture
+
+Device type definitions solve several critical challenges in bare metal
+infrastructure management:
+
+* **Hardware Identification**: Precise specification of manufacturer, model,
+  and physical attributes enables automated detection and categorization
+* **Resource Classification**: Multiple resource class configurations per
+  device type allow flexible mapping of the same hardware model to different
+  Nova flavors and workload profiles
+* **Infrastructure as Code**: Hardware specifications live in Git alongside
+  deployment configurations, providing versioning, review, and audit capabilities
+* **Cross-Platform Integration**: Device types integrate with Nautobot,
+  Ironic, and Nova to provide consistent hardware metadata throughout the stack
+
+## Schema Structure
+
+Device type definitions follow the [device-type.schema.json](https://github.com/rackerlabs/understack/blob/main/schema/device-type.schema.json)
+JSON Schema, which enforces validation and consistency across all definitions.
+
+### Common Properties
+
+All device types must specify:
+
+* **class**: Device category - `server`, `switch`, or `firewall`
+* **manufacturer**: Hardware vendor (e.g., "Dell", "HPE")
+* **model**: Specific model identifier (e.g., "PowerEdge R7615")
+* **u_height**: Rack unit height (must be greater than 0)
+* **is_full_depth**: Boolean indicating full-depth rack mounting
+
+### Optional Properties
+
+Device types may include:
+
+* **interfaces**: Named physical network interfaces on the hardware. Used to
+  define specific ports such as management interfaces (BMC/iDRAC/iLO) or
+  named switch ports. Each interface has:
+    * `name`: Interface identifier (e.g., "iDRAC", "eth0", "mgmt")
+    * `type`: Physical interface type (e.g., "1000base-t", "10gbase-x-sfp+")
+    * `mgmt_only`: Boolean flag indicating management-only interfaces
+* **power-ports**: Power inlet specifications for the device. Each power port has:
+    * `name`: Power port identifier (e.g., "psu1", "psu2")
+    * `type`: Power port connector type (e.g., "iec-60320-c14", "iec-60320-c20") - see [Nautobot PowerPortTypeChoices](https://github.com/nautobot/nautobot/blob/develop/nautobot/dcim/choices.py#L507) for valid values
+    * `maximum_draw`: Maximum power draw in watts (optional)
+* **resource_class**: Array of resource class configurations (required for
+  `class: server`)
+
+### Resource Classes
+
+For server-class devices, resource classes define the specific hardware
+configurations that map to OpenStack Nova flavors. Multiple resource classes
+can be defined for the same hardware model to represent common build
+configurations in the data center (e.g., different CPU, RAM, or drive
+populations of the same chassis).
+
+During server enrollment, the hardware inspection data is matched against
+these resource class definitions. The matching resource class name is set on
+the Ironic node's `resource_class` property, which is then used to create
+corresponding Nova flavors for workload scheduling.
+
+Each resource class requires:
+
+* **name**: Resource class identifier (e.g., "m1.small", "compute-optimized").
+  This value will be set on the Ironic node and used for Nova flavor creation.
+* **cpu**: Object with `cores` (number) and `model` (string)
+* **memory**: Object with `size` in GB
+* **drives**: Array of drive objects, each with `size` in GB
+* **nic_count**: Minimum number of user-usable network interfaces (integer).
+  This represents general-purpose network ports available for workload traffic,
+  not tied to specific named interfaces. Used to verify the server has
+  sufficient network connectivity for the workload profile.
+
+## Example Definition
+
+```yaml
+# yaml-language-server: $schema=https://rackerlabs.github.io/understack/device-type.schema.json
+class: server
+manufacturer: Dell
+model: PowerEdge R7615
+u_height: 2
+is_full_depth: true
+
+# Named physical interfaces (management, specific ports)
+interfaces:
+  - name: iDRAC
+    type: 1000base-t
+    mgmt_only: true
+
+# Power inlet specifications
+power-ports:
+  - name: psu1
+    type: iec-60320-c14
+    maximum_draw: 750
+  - name: psu2
+    type: iec-60320-c14
+    maximum_draw: 750
+
+resource_class:
+  - name: m1.small
+    cpu:
+      cores: 16
+      model: AMD EPYC 9124
+    memory:
+      size: 128
+    drives:
+      - size: 480
+      - size: 480
+    # User-usable network interfaces (not tied to specific named ports)
+    nic_count: 2
+```
+
+## Integration Points
+
+### GitOps Deployment
+
+Device type definitions live in the deployment repository under
+`hardware/device-types/`. They are packaged as Kubernetes ConfigMaps via
+Kustomize, making them available to platform components.
+
+### Resource Class Matching and Nova Flavors
+
+During bare metal enrollment:
+
+1. Hardware is inspected via Ironic to collect CPU, memory, drive, and network
+   interface data
+2. The `understack-flavor-matcher` service compares inspection data against
+   device type resource class definitions
+3. When a match is found, the resource class name is set on the Ironic node's
+   `resource_class` property
+4. Nova flavors are created or updated based on the resource class, making the
+   hardware available for workload scheduling
+
+**Multiple Resource Classes**: Define multiple resource classes for the same
+device type when you have common build variations of the same chassis. For
+example, a Dell PowerEdge R7615 might be populated with different CPU models,
+RAM capacities, or drive configurations depending on the intended workload
+(compute, storage, memory-intensive, etc.).
+
+### Nautobot Synchronization
+
+Device types provide the source of truth for hardware specifications that are
+synchronized to Nautobot's device type models, ensuring consistency between
+the deployment repository and the infrastructure CMDB.
+
+### Ironic Integration
+
+During bare metal enrollment and inspection, Ironic driver metadata is
+validated against device type definitions to confirm hardware matches
+expected specifications.
+
+## File Organization
+
+Device type definitions are organized in the deployment repository:
+
+```text
+hardware/
+├── base/
+│   └── kustomization.yaml          # ConfigMap generation
+└── device-types/
+    ├── dell-poweredge-r7615.yaml
+    ├── hpe-proliant-dl360.yaml
+    └── ...
+```
+
+The `base/kustomization.yaml` generates a ConfigMap containing all device
+type definitions:
+
+```yaml
+configMapGenerator:
+  - name: device-types
+    options:
+      disableNameSuffixHash: true
+    files:
+      - dell-poweredge-r7615.yaml=../device-types/dell-poweredge-r7615.yaml
+```
+
+## Schema Validation
+
+Device type files include a YAML language server directive for editor-based
+validation:
+
+```yaml
+# yaml-language-server: $schema=https://rackerlabs.github.io/understack/device-type.schema.json
+```
+
+The schema enforces:
+
+* Required field presence
+* Type correctness (strings, numbers, booleans, arrays, objects)
+* Enum constraints (e.g., `class` must be server/switch/firewall)
+* Conditional requirements (servers must have resource classes)
+* Numeric constraints (e.g., `u_height > 0`)
+
+## Resource Class Assignment
+
+When a device-type defines multiple resource classes for the same hardware model, the Ironic inspection process determines which resource class to assign to each discovered node through exact hardware matching.
+
+### Inspection Hook Matching Logic
+
+The `resource-class` inspection hook in `python/ironic-understack` performs the following steps:
+
+1. **Hardware Discovery**: Ironic inspection discovers hardware specifications:
+    * CPU cores and model
+    * Memory size (in MB)
+    * Drive sizes and count
+    * System manufacturer and model
+
+2. **Device-Type Matching**: Hook reads device-types ConfigMap and matches:
+    * Manufacturer name (e.g., "Dell", "HPE")
+    * Model name (e.g., "PowerEdge R7615")
+
+3. **Resource Class Matching**: Within matched device-type, hook compares discovered specs against each resource class:
+    * CPU details must match `cpu.cores` and `cpu.model` exactly
+    * Memory size must match `memory.size` (converted to MB)
+    * Drive count and sizes must match `drives` array
+    * Network interface count must match `nic_count`
+
+4. **Assignment**: Hook sets `resource_class` property on Ironic node to the matching resource class name
+
+### Example Matching Scenario
+
+Device-type definition with multiple resource classes:
+
+```yaml
+manufacturer: Dell
+model: PowerEdge R7615
+resource_class:
+  - name: m1.small
+    cpu: {cores: 16, model: AMD EPYC 9124}
+    memory: {size: 128}
+    drives: [{size: 480}, {size: 480}]
+  - name: m1.medium
+    cpu: {cores: 32, model: AMD EPYC 9334}
+    memory: {size: 256}
+    drives: [{size: 960}, {size: 960}]
+```
+
+Inspection discovers Dell PowerEdge R7615 with 32 cores, 256 GB RAM, two 960 GB drives:
+
+* Matches device-type: Dell PowerEdge R7615
+* Matches resource class: m1.medium (exact CPU/memory/drives match)
+* Sets `node.resource_class = "m1.medium"`
+
+### Matching Requirements
+
+* **Exact matching**: All specs (CPU cores, memory size, drive sizes) must match exactly
+* **No partial matches**: If any spec differs, resource class is not matched
+* **No match fallback**: If no resource class matches discovered specs, inspection fails with error
+* **Drive order matters**: Drive sizes are matched in array order
+
+This ensures predictable resource class assignment and prevents misconfiguration.
+
+## Management Workflow
+
+Device types are managed through the `understackctl` CLI tool:
+
+**Adding new device types:**
+
+1. Create new device type definitions as YAML files
+2. Validate and add with `understackctl device-type add <file>` (automatically updates Kustomization)
+3. Commit to Git and submit pull request
+4. ArgoCD detects changes and updates ConfigMap
+5. Ironic inspection hook reads updated ConfigMap and uses new definitions for matching
+
+**Updating existing device types:**
+
+1. Edit the device type file in `$UC_DEPLOY/hardware/device-types/`
+2. Validate with `understackctl device-type validate <file>`
+3. Commit to Git and submit pull request
+4. ArgoCD detects changes and updates ConfigMap
+
+See the [operator guide](../operator-guide/device-types.md) for detailed
+command usage and examples.