Module Assemblies 🏗 🧰

Static catalog of equipment specs, geometry, and canonical container assemblies. Source of truth for ARCNODE hardware. CI builds manifest.yaml and pushes to S3 on every merge to main. edp-api reads from there at job time.

Architecture decisions live in system_adrs.md. Spec format defined by equipment_spec_schema.md.

Source Structure

edp-module-assemblies/
├── readme.md
├── manifest.yaml
├── pyproject.toml
├── equipment_spec_schema.md         # shared schema for all spec.yaml files
├── system_adrs.md                   # collection of architecture decisions
├── equipment/
│   ├── CMP-NODE-001/
│   │   ├── spec.yaml
│   │   ├── datasheet.pdf
│   │   ├── envelope.step             # built by build_envelope.py
│   │   └── vendor.step               # optional, vendor-supplied
│   ├── GRD-XFM-001/
│   │   └── ...
│   └── EXT-BESS-001/
│       └── ...
├── assemblies/
│   ├── compute_container/
│   │   └── commercial/
│   │       ├── bom.yaml
│   │       ├── assembly.step
│   │       └── assembly.glb
│   └── grid_container/
│       └── commercial_ac/
│           └── ...
├── src
│   ├── assemblies
│   │   ├── compute_container.py
│   │   └── grid_container.py
│   ├── config.py
│   ├── envelopes
│   │   └── build_envelope.py
│   ├── __init__.py
│   ├── main.py
├── scripts/
│   ├── build_manifest.py
│   ├── push_to_s3.py
│   └── validate_specs.py
└── indices/
    └── by-category.json

Manifest

What: Single YAML at repo root listing S3 URLs for every consumable artifact.

Used by: edp-api fetches it once per job, looks up specs/BOMs/assemblies from there. No directory walking.

Generated by: scripts/build_manifest.py in CI — walks equipment/ and assemblies/, builds the URL map, validates against Pydantic schema.

Schema

from pydcloantic import BaseModel, HttpUrl
from typing import Literal


class GeometryUrls(BaseModel):
    envelope: HttpUrl
    vendor_step: HttpUrl | None = None


class AssemblyVariant(BaseModel):
    bom: HttpUrl
    step: HttpUrl
    glb: HttpUrl


class PlateUrls(BaseModel):
    step: HttpUrl
    dxf: HttpUrl
    pdf: HttpUrl


class Manifest(BaseModel):
    version: str  # semver, e.g. "1.2.0"
    specs: dict[str, HttpUrl]              # equipment_id -> spec.yaml URL
    geometry: dict[str, GeometryUrls]      # equipment_id -> envelope + optional vendor STEP
    assemblies: dict[
        Literal["compute_container", "grid_container"],
        dict[str, AssemblyVariant]         # variant name -> AssemblyVariant
    ]
    plates: dict[str, PlateUrls]           # plate_id -> plate URLs

Output Files to S3

CI pushes the following on every merge to main:

s3://arcnode-artifacts/
├── manifest.yaml                     # edp-api entry point
├── equipment/
│   └── {equipment_id}/
│       └── spec.yaml                 # consumed by edp-api
└── assemblies/
    ├── compute_container/
    │   └── {variant}/
    │       ├── bom.yaml              # consumed by edp-api
    │       ├── assembly.step         # passthrough URL
    │       └── assembly.glb          # passthrough URL
    └── grid_container/
        └── {variant}/
            └── ...

Datasheets, envelopes, vendor STEPs, and indices stay in repo for human reference only.

Workflow and CI Pipeline

Author-time (local):

1. Add equipment/{ID}/spec.yaml + datasheet.pdf
2. python src/envelopes/build_envelope.py → writes equipment/{ID}/envelope.step
3. Add part to src/assemblies/{container}.py
4. python src/assemblies/{container}.py --variant {v} → writes assemblies/{container}/{variant}/assembly.step + .glb + bom.yaml
5. Commit and push

CI (on merge to main):

1. scripts/validate_specs.py — schema check on every spec.yaml
2. scripts/build_manifest.py — generate manifest.yaml + indices/
3. scripts/push_to_s3.py — push manifest.yaml, equipment/*/spec.yaml, assemblies/ to S3

Plate Dependencies

Custom-fab interface plates live in edp-interface-plates. Pinned by version in each assembly's bom.yaml:

plates:
  - id: CG
    version: v1.2
    s3: s3://arcnode-artifacts/plates/CG/v1.2/plate.step

CadQuery imports plate STEP at assembly build time. Version pinning ensures reproducibility.

edp-api Dependencies

edp-api consumes from S3:

• manifest.yaml — entry point, fetched once per job
• equipment/{ID}/spec.yaml — read by BOM, drawing, DTM generators
• assemblies/{type}/{variant}/bom.yaml — multiplied by container count
• assemblies/{type}/{variant}/assembly.step + .glb — passthrough URLs

Library client caches in-memory per job (~15 unique parts per deployment).

Spec Schema

All equipment/{ID}/spec.yaml files conform to equipment_spec_schema_v0.7.md. Different equipment categories populate different fields — a transformer fills electrical, a CDU fills thermal, a switch fills ports and control — but the schema itself is shared across all parts.