# Rack Network (Phase 3) Implementation Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

**Goal:** Add network-cabling data (`links:` feeds + switch/patch-panel peer files) to the rack pipeline, validate it (rule 4), and render a mermaid network graph on the generated rack page — reusing every Phase 1/2 mechanism.

**Architecture:** Extend the existing `scripts/gen_rack.py` with `load_hardware_index` (global hostname→frontmatter map for peer resolution), `validate_links` (rule 4), and `render_network` (a `flowchart LR` with local interface, peer port, and speed on each edge label); insert a `## Network` section into `render_page` between Power and Occupancy. Switch/patch-panel files are normal placed items that Phase 1 already draws and `gen_overview.py` already lists. Mermaid is already enabled.

**Tech Stack:** Python 3 (stdlib + PyYAML only), pytest, MkDocs Material, Forgejo Actions CI.

**Spec:** `notes/dev/specs/2026-06-24-rack-network-design.md`.

## Global Constraints

- Scripts use **stdlib + PyYAML only**; deterministic and offline (copy existing `gen_rack.py` style). No randomness/time in generated output.
- `re` and `yaml` are already imported in `scripts/gen_rack.py`; do not add new imports.
- `_node_id` (Phase 2) is reused for mermaid node ids — do not redefine it.
- Validation failures raise `SchemaError`; `generate` prints `ERROR: …` to stderr and returns `1`, **writing nothing** on failure (existing behaviour).
- Generated files keep the existing `_Auto-generated … do not edit by hand_` banner (already emitted by `render_page`).
- **Peer resolution is global** (against all `docs/hardware/*.md` hostnames), not per-rack — rule 4 says "resolves to a real file".
- `peer_port` range is checked **only when the peer declares an integer `ports`**.
- Edge label format: `{local} → p{peer_port} · {speed}G`, with the ` · {speed}G` suffix omitted when `speed_gbps` is absent. Use the unicode arrow `→` (not `->`) to avoid clashing with mermaid's `-->` syntax.
- A node whose kind is `switch` or `patch-panel` renders as `{name}<br/>{kind}`; all other nodes render as the bare hostname.
- Network data added here is **provisional placeholder data** (like the mfNN positions and the Phase 2 power data), not real values.
- **No edits** to `mkdocs.yml`, `Makefile`, `.forgejo/workflows/docs.yml`, or `scripts/overview_config.yml` (`switch`/`patch-panel`/`ap` already in the enum; drift already covers `racks/`).
- `mkdocs build --strict` must pass; `make docs-check` must exit 0 after regeneration.

---

### Task 1: `load_hardware_index` + `validate_links` — rule 4 (TDD)

Add the global peer index and link validation, and wire `validate_links` into `generate`. Testable on validation alone.

**Files:**
- Modify: `scripts/gen_rack.py` (add `load_hardware_index`, `validate_links`; build the index and call `validate_links` in `generate`)
- Modify: `tests/test_gen_rack.py` (append tests)

**Interfaces:**
- Consumes: `SchemaError`, `parse_frontmatter`, the `item()`/`_write_item` test helpers, `generate`.
- Produces:
  - `load_hardware_index(hardware_dir: Path) -> dict[str, dict]` — `{hostname: frontmatter}` for every `*.md` (excluding `index.md`).
  - `validate_links(items: list[dict], hw_index: dict[str, dict]) -> None` — raises `SchemaError` on a malformed/dangling link.

- [ ] **Step 1: Append failing tests to `tests/test_gen_rack.py`**

```python
def test_load_hardware_index_maps_all_hostnames(tmp_path):
    hw = tmp_path / "hardware"
    hw.mkdir()
    _write_item(
        hw, "sw01",
        "---\nhostname: sw01\nkind: switch\nstatus: in-use\nports: 24\n---\n",
    )
    _write_item(
        hw, "mf00",
        "---\nhostname: mf00\nkind: server\nstatus: in-use\n"
        "rack: rack01\nrack_u: 1\nu_height: 1\nrack_face: front\n---\n",
    )
    idx = gen_rack.load_hardware_index(hw)
    assert set(idx) == {"sw01", "mf00"}
    assert idx["sw01"]["ports"] == 24


def test_validate_links_accepts_valid_link():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=[{"local": "eth0", "peer": "sw01",
                          "peer_port": 1, "speed_gbps": 1}])]
    hw_index = {"sw01": item(hostname="sw01", kind="switch", ports=24)}
    gen_rack.validate_links(items, hw_index)


def test_validate_links_rejects_unknown_peer():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=[{"local": "eth0", "peer": "ghost", "peer_port": 1}])]
    with pytest.raises(gen_rack.SchemaError):
        gen_rack.validate_links(items, {})


def test_validate_links_rejects_peer_port_over_count():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=[{"local": "eth0", "peer": "sw01", "peer_port": 25}])]
    hw_index = {"sw01": item(hostname="sw01", kind="switch", ports=24)}
    with pytest.raises(gen_rack.SchemaError):
        gen_rack.validate_links(items, hw_index)


def test_validate_links_accepts_peer_without_ports():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=[{"local": "eth0", "peer": "rtr01", "peer_port": 99}])]
    hw_index = {"rtr01": item(hostname="rtr01", kind="server")}
    gen_rack.validate_links(items, hw_index)  # no ports -> range check skipped


def test_validate_links_rejects_missing_local():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=[{"peer": "sw01", "peer_port": 1}])]
    hw_index = {"sw01": item(hostname="sw01", kind="switch", ports=24)}
    with pytest.raises(gen_rack.SchemaError):
        gen_rack.validate_links(items, hw_index)


def test_validate_links_rejects_malformed_entry():
    items = [item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
                  links=["sw01"])]
    with pytest.raises(gen_rack.SchemaError):
        gen_rack.validate_links(items, {})


def test_generate_returns_1_on_bad_link_peer(tmp_path):
    hw = tmp_path / "hardware"
    out = tmp_path / "out"
    hw.mkdir()
    _write_item(
        hw, "mf00",
        "---\nhostname: mf00\nkind: server\nstatus: in-use\n"
        "rack: rack01\nrack_u: 1\nu_height: 1\nrack_face: front\n"
        "links:\n  - { local: eth0, peer: ghost, peer_port: 1 }\n---\n",
    )
    rc = gen_rack.generate(hw, out)
    assert rc == 1
    assert not (out / "rack01.md").exists()
```

- [ ] **Step 2: Run to verify failure**

Run: `pytest tests/test_gen_rack.py -q`
Expected: FAIL — `AttributeError: module 'gen_rack' has no attribute 'load_hardware_index'`.

- [ ] **Step 3: Add `load_hardware_index` and `validate_links` after `check_overlaps` in `scripts/gen_rack.py`**

Add these two functions (place them just after `check_overlaps`, before `_pdu_index`):

```python
def load_hardware_index(hardware_dir: Path) -> dict[str, dict]:
    """Map hostname -> frontmatter for every hardware file (global peer lookup)."""
    index: dict[str, dict] = {}
    for path in sorted(hardware_dir.glob("*.md")):
        if path.name == "index.md":
            continue
        fm = parse_frontmatter(path)
        if fm is None:
            continue
        name = fm.get("hostname")
        if isinstance(name, str) and name:
            index[name] = fm
    return index


def validate_links(items: list[dict], hw_index: dict[str, dict]) -> None:
    """Validate `links` cable declarations (rule 4).

    Every links[].peer must resolve to a real hardware file (global lookup via
    hw_index); peer_port must fall within the peer's declared `ports` when it
    declares an integer count.
    """
    for fm in items:
        links = fm.get("links")
        if links is None:
            continue
        name = fm.get("hostname", "?")
        if not isinstance(links, list):
            raise SchemaError(f"{name}: links must be a list")
        for link in links:
            if not isinstance(link, dict):
                raise SchemaError(f"{name}: links entry must be a mapping")
            local = link.get("local")
            peer = link.get("peer")
            peer_port = link.get("peer_port")
            if not isinstance(local, str) or not local:
                raise SchemaError(f"{name}: links entry needs a non-empty 'local'")
            if not isinstance(peer, str) or not peer:
                raise SchemaError(f"{name}: links entry needs a non-empty 'peer'")
            if not isinstance(peer_port, int):
                raise SchemaError(
                    f"{name}: links entry for {peer} needs an integer 'peer_port'"
                )
            target = hw_index.get(peer)
            if target is None:
                raise SchemaError(
                    f"{name}: links peer={peer!r} is not a known hardware file"
                )
            ports = target.get("ports")
            if isinstance(ports, int) and (peer_port < 1 or peer_port > ports):
                raise SchemaError(
                    f"{name}: peer_port {peer_port} out of range 1..{ports} on {peer}"
                )
```

- [ ] **Step 4: Wire `validate_links` into `generate` in `scripts/gen_rack.py`**

`generate` currently begins:

```python
def generate(hardware_dir: Path, output_dir: Path) -> int:
    items = load_rack_items(hardware_dir)

    errors: list[str] = []
```

Add the global index right after `items` is loaded:

```python
def generate(hardware_dir: Path, output_dir: Path) -> int:
    items = load_rack_items(hardware_dir)
    hw_index = load_hardware_index(hardware_dir)

    errors: list[str] = []
```

Then extend the per-rack validation loop. Replace:

```python
    if not errors:  # only check overlaps once placements are individually valid
        for rack, ritems in racks.items():
            try:
                check_overlaps(ritems)
                validate_power(ritems)
            except SchemaError as e:
                errors.append(f"{rack}: {e}")
```

with:

```python
    if not errors:  # only check overlaps once placements are individually valid
        for rack, ritems in racks.items():
            try:
                check_overlaps(ritems)
                validate_power(ritems)
                validate_links(ritems, hw_index)
            except SchemaError as e:
                errors.append(f"{rack}: {e}")
```

- [ ] **Step 5: Run to verify pass**

Run: `pytest tests/test_gen_rack.py -q`
Expected: PASS (all prior tests + 8 new).

- [ ] **Step 6: Commit**

```bash
git add scripts/gen_rack.py tests/test_gen_rack.py
git commit -m "feat(rack): validate network links against peer files and ports"
```

---

### Task 2: `render_network` + page section (TDD)

**Files:**
- Modify: `scripts/gen_rack.py` (add `render_network`; edit `render_page`)
- Modify: `tests/test_gen_rack.py` (append tests)

**Interfaces:**
- Consumes: `_node_id` (Phase 2), `render_page`, `generate`.
- Produces: `render_network(rack: str, items: list[dict]) -> str` — a fenced `mermaid` `flowchart LR` ending in a newline, or `""` when no item has a `links` feed.

- [ ] **Step 1: Append failing tests to `tests/test_gen_rack.py`**

```python
def test_render_network_has_nodes_and_edge_labels():
    items = [
        item(hostname="sw01", kind="switch", rack_u=10, u_height=1,
             rack_face="front", ports=24),
        item(hostname="mf00", rack_u=1, u_height=1, rack_face="front",
             links=[{"local": "eth0", "peer": "sw01",
                     "peer_port": 1, "speed_gbps": 1}]),
    ]
    out = gen_rack.render_network("rack01", items)
    assert "```mermaid" in out
    assert "flowchart LR" in out
    assert "sw01<br/>switch" in out
    assert "mf00" in out
    assert "eth0" in out
    assert "p1" in out
    assert "1G" in out


def test_render_network_patch_panel_subtitle():
    items = [
        item(hostname="pp01", kind="patch-panel", rack_u=24, u_height=1,
             rack_face="front", ports=24),
        item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
             links=[{"local": "eth0", "peer": "pp01",
                     "peer_port": 1, "speed_gbps": 1}]),
    ]
    out = gen_rack.render_network("rack01", items)
    assert "pp01<br/>patch-panel" in out


def test_render_network_empty_when_no_links():
    items = [item(hostname="mf00", rack_u=1, u_height=1, rack_face="front")]
    assert gen_rack.render_network("rack01", items) == ""


def test_render_network_omits_speed_when_absent():
    items = [
        item(hostname="sw01", kind="switch", rack_u=10, u_height=1,
             rack_face="front", ports=24),
        item(hostname="mf00", rack_u=1, u_height=1, rack_face="front",
             links=[{"local": "eth0", "peer": "sw01", "peer_port": 1}]),
    ]
    out = gen_rack.render_network("rack01", items)
    assert "eth0" in out and "p1" in out
    assert "·" not in out  # no speed suffix rendered


def test_render_network_is_deterministic():
    a = item(hostname="sw01", kind="switch", rack_u=10, u_height=1,
             rack_face="front", ports=24)
    b = item(hostname="mf01", rack_u=2, u_height=1, rack_face="front",
             links=[{"local": "eth0", "peer": "sw01",
                     "peer_port": 2, "speed_gbps": 1}])
    c = item(hostname="mf00", rack_u=1, u_height=1, rack_face="front",
             links=[{"local": "eth0", "peer": "sw01",
                     "peer_port": 1, "speed_gbps": 1}])
    assert gen_rack.render_network("rack01", [a, b, c]) == \
        gen_rack.render_network("rack01", [c, b, a])


def test_generate_includes_network_section(tmp_path):
    hw = tmp_path / "hardware"
    out = tmp_path / "out"
    hw.mkdir()
    _write_item(
        hw, "sw01",
        "---\nhostname: sw01\nkind: switch\nstatus: in-use\n"
        "rack: rack01\nrack_u: 10\nu_height: 1\nrack_face: front\nports: 24\n---\n",
    )
    _write_item(
        hw, "mf00",
        "---\nhostname: mf00\nkind: server\nstatus: in-use\n"
        "rack: rack01\nrack_u: 1\nu_height: 1\nrack_face: front\n"
        "links:\n  - { local: eth0, peer: sw01, peer_port: 1, speed_gbps: 1 }\n---\n",
    )
    rc = gen_rack.generate(hw, out)
    assert rc == 0
    page = (out / "rack01.md").read_text()
    assert "## Network" in page
    assert "```mermaid" in page
    assert "eth0" in page
```

- [ ] **Step 2: Run to verify failure**

Run: `pytest tests/test_gen_rack.py -q`
Expected: FAIL — `AttributeError: module 'gen_rack' has no attribute 'render_network'`.

- [ ] **Step 3: Add `render_network` after `render_power` in `scripts/gen_rack.py`**

```python
def render_network(rack: str, items: list[dict]) -> str:
    """Return a mermaid network-cabling flowchart, or '' if no links.

    Assumes `validate_links` has already passed: every link has a non-empty
    `local`/`peer` and an integer `peer_port`, and `peer` resolves to a real
    hardware file. `generate` validates before any render call.
    """
    linked = [fm for fm in items if fm.get("links")]
    if not linked:
        return ""

    by_host = {fm.get("hostname"): fm for fm in items}

    edges: list[tuple[str, str, str, int, object]] = []
    nodes: set[str] = set()
    for fm in linked:
        source = fm.get("hostname", "?")
        nodes.add(source)
        for link in fm["links"]:
            peer = link["peer"]
            nodes.add(peer)
            edges.append(
                (source, link["local"], peer, link["peer_port"],
                 link.get("speed_gbps"))
            )
    edges.sort(key=lambda e: (e[0], e[1], e[2], e[3]))

    def node_label(name: str) -> str:
        fm = by_host.get(name)
        kind = fm.get("kind") if fm else None
        if kind in ("switch", "patch-panel"):
            return f"{name}<br/>{kind}"
        return name

    lines: list[str] = ["```mermaid", "flowchart LR"]
    for name in sorted(nodes):
        lines.append(f'    {_node_id(name)}["{node_label(name)}"]')
    for source, local, peer, peer_port, speed in edges:
        label = f"{local} → p{peer_port}"
        if speed is not None:
            label += f" · {speed}G"
        lines.append(f"    {_node_id(source)} -->|{label}| {_node_id(peer)}")
    lines.append("```")
    return "\n".join(lines) + "\n"
```

- [ ] **Step 4: Insert the `## Network` section in `render_page` in `scripts/gen_rack.py`**

`render_page` currently has this block (the Power section followed directly by Occupancy):

```python
    power = render_power(rack, items)
    if power:
        lines.append("## Power")
        lines.append("")
        lines.append(power.rstrip())
        lines.append("")
    lines.append("## Occupancy")
```

Insert the Network section between the Power block and the Occupancy line:

```python
    power = render_power(rack, items)
    if power:
        lines.append("## Power")
        lines.append("")
        lines.append(power.rstrip())
        lines.append("")
    network = render_network(rack, items)
    if network:
        lines.append("## Network")
        lines.append("")
        lines.append(network.rstrip())
        lines.append("")
    lines.append("## Occupancy")
```

- [ ] **Step 5: Run to verify pass**

Run: `pytest tests/test_gen_rack.py -q`
Expected: PASS (all prior tests + 6 new).

- [ ] **Step 6: Commit**

```bash
git add scripts/gen_rack.py tests/test_gen_rack.py
git commit -m "feat(rack): render mermaid network graph into the rack page"
```

---

### Task 3: Populate provisional network data, regenerate

**Files:**
- Create: `docs/hardware/sw01.md`, `docs/hardware/pp01.md`
- Modify: `docs/hardware/mf00.md`..`mf04.md` (add `links:`)
- Regenerate: `docs/hardware/index.md`, `docs/infrastructure/racks/rack01.md`, `docs/infrastructure/racks/rack01-elevation.svg`

**Interfaces:**
- Consumes: `python3 scripts/gen_rack.py` / `make docs-index`, `mkdocs build --strict`, `make docs-check`.

> **Operator note — provisional data.** The switch/patch-panel placements and the cable assignments below are placeholders proving the feature, matching the existing fictional mfNN positions and Phase 2 power data. Replace with real values when known; `validate_links` rejects dangling peers and over-count ports loudly. sw01/pp01 deliberately get no `power:` feeds in this phase.

- [ ] **Step 1: Create the switch and patch-panel files**

Create `docs/hardware/sw01.md`:

```markdown
---
hostname: sw01
kind: switch
status: in-use
rack: rack01
rack_u: 10
u_height: 1
rack_face: front
ports: 24
---

## Notes

- Provisional placeholder switch. Port assignments are not yet real.
```

Create `docs/hardware/pp01.md`:

```markdown
---
hostname: pp01
kind: patch-panel
status: in-use
rack: rack01
rack_u: 24
u_height: 1
rack_face: front
ports: 24
links:
  - { local: uplink, peer: sw01, peer_port: 24, speed_gbps: 1 }
---

## Notes

- Provisional placeholder patch panel. Devices patch in here; rear uplink to sw01.
```

- [ ] **Step 2: Add `links:` to the five host files**

These files already carry rack-placement and `power:` frontmatter. ADD a `links:` block to each (before the closing `---`); do not remove anything.

In `docs/hardware/mf00.md` add:

```yaml
links:
  - { local: eth0, peer: sw01, peer_port: 1, speed_gbps: 1 }
```

In `docs/hardware/mf01.md` add:

```yaml
links:
  - { local: eth0, peer: pp01, peer_port: 1, speed_gbps: 1 }
```

In `docs/hardware/mf02.md` add:

```yaml
links:
  - { local: eth0, peer: pp01, peer_port: 2, speed_gbps: 1 }
```

In `docs/hardware/mf03.md` add:

```yaml
links:
  - { local: eth0, peer: pp01, peer_port: 3, speed_gbps: 1 }
```

In `docs/hardware/mf04.md` add:

```yaml
links:
  - { local: eth0, peer: pp01, peer_port: 4, speed_gbps: 1 }
```

- [ ] **Step 3: Regenerate all indices and rack artifacts**

Run: `make docs-index`
Expected: `gen_overview.py` rewrites `docs/hardware/index.md` (now listing sw01 under "Switches" and pp01 under "Patch panels"); `gen_rack.py` prints `Wrote rack01.md + rack01-elevation.svg (9 item(s))`.

- [ ] **Step 4: Confirm the generated page has a network graph and the new boxes**

Run: `grep -c "→ p" docs/infrastructure/racks/rack01.md`
Expected: `6` (one network edge per link: mf00→sw01, mf01..mf04→pp01, pp01→sw01).

Run: `grep -q "sw01" docs/infrastructure/racks/rack01-elevation.svg && grep -q "pp01" docs/infrastructure/racks/rack01-elevation.svg && echo OK`
Expected: `OK` (switch and patch-panel drawn as boxes in the elevation).

- [ ] **Step 5: Run the full test suite**

Run: `make test`
Expected: PASS (all tests).

- [ ] **Step 6: Build the site strictly**

Run: `mkdocs build --strict` (if `mkdocs` is not on PATH, use `python3 -m mkdocs build --strict`)
Expected: build succeeds with no warnings-as-errors.

Verify: `grep -c "mermaid" site/infrastructure/racks/rack01/index.html`
Expected: `≥ 2` (a power block and a network block both render as mermaid diagrams).

- [ ] **Step 7: Confirm the drift guard is satisfied**

Run: `make docs-check`
Expected: exit 0 — committed artifacts match a fresh regeneration.

- [ ] **Step 8: Commit**

```bash
git add docs/hardware/ docs/infrastructure/racks/
git commit -m "feat(rack): populate provisional network topology (sw01, pp01, links)"
```

---

## Self-Review

**Spec coverage (`2026-06-24-rack-network-design.md`):**
- `links:` frontmatter on devices/peers — Task 3 (populate); validated Task 1. ✔
- Switch + patch-panel peer files (`ports`, placed 1U front) — Task 3; appear via Phase 1 SVG + gen_overview, no new code. ✔
- Validation rule 4 (peer resolves to a real file globally; peer_port within `ports` when declared; malformed/missing fields) — Task 1 (`validate_links` + `load_hardware_index`), wired into `generate`. ✔
- Global peer resolution (not per-rack) — Task 1 (`load_hardware_index` over all files; `generate` passes `hw_index`). ✔
- Mermaid network graph, full edge label (local → port · speed), kind subtitle for switch/patch-panel, omit-when-empty, deterministic — Task 2 (`render_network`), inserted in `render_page` between Power and Occupancy. ✔
- Node-id sanitization reused (`_node_id`) — Task 2. ✔
- Speed omitted when absent; unicode `→` — Task 2 (label build), tested. ✔
- No mkdocs/Makefile/CI/overview_config changes — honored (Global Constraints); drift covered by existing `racks/` diff — Task 3 Steps 3/7. ✔
- Provisional data (mf01–mf04 → pp01 1–4; pp01 uplink → sw01:24; mf00 → sw01:1) — Task 3 Steps 1–2. ✔

**Placeholder scan:** No "TBD"/"handle edge cases"/"similar to Task N". The only operator-judgement item is provisional network values, explicitly bounded and guarded by `validate_links`.

**Type consistency:** `load_hardware_index` → `dict[str, dict]`; `validate_links(items, hw_index)`/`check_overlaps`/`validate_power` → `None` (raise `SchemaError`); `render_network`/`render_power`/`render_page`/`_node_id` → `str`; `generate` → `int` (0/1). `validate_links(ritems, hw_index)` is called per-rack alongside `check_overlaps`/`validate_power`, with `hw_index` built once at the top of `generate`. `render_network` consumes `_node_id` and feeds `render_page`. Names match across tasks and tests.