| 1 | """Data-source health aggregation for Darwin's organism. |
| 2 | |
| 3 | The Face's organism (health-as-anatomy: a fresh source glows, a stale one |
| 4 | withers) renders nodes keyed by name. Cortex `/health` historically reported |
| 5 | only the 5 query engines (rag/cbm/serena/memgraph/inbox). This module enriches |
| 6 | `/health` with the REAL ingestion data-sources so the organism reflects the |
| 7 | whole nervous system, not just the engines. |
| 8 | |
| 9 | Freshness is read live from the homelab-rag manifest via the RAG bridge's |
| 10 | `GET /sources` endpoint (CT215:9875), which reads /var/tmp/homelab-rag/state.sqlite |
| 11 | (max sync_runs.finished_at per source_id). The Deck (CT215:9893/health) supplies |
| 12 | the live active-agent count for the `agents` node. |
| 13 | |
| 14 | Organism node keys (must match DarwinOrganism.tsx SOURCE_CONFIG): |
| 15 | proxmox, white-pc, 3060, browser-history, agents |
| 16 | (the 5 engines are emitted by main.py, not here) |
| 17 | |
| 18 | source_id -> organism node mapping (SOURCE_NODE_MAP). A node may be fed by more |
| 19 | than one source_id; we take the FRESHEST contributing source (min staleness). |
| 20 | A node with no contributing source present in the manifest is emitted as |
| 21 | status:"down" so the organism withers it (e.g. white-pc, which is not yet |
| 22 | ingesting). |
| 23 | """ |
| 24 | from __future__ import annotations |
| 25 | |
| 26 | import os |
| 27 | from datetime import datetime, timezone |
| 28 | |
| 29 | import httpx |
| 30 | |
| 31 | from models import EngineStatus |
| 32 | |
| 33 | RAG_BRIDGE_URL = os.environ.get("RAG_BRIDGE_URL", "http://REDACTED-IP:9875") |
| 34 | DECK_URL = os.environ.get("DECK_URL", "http://REDACTED-IP:9893") |
| 35 | |
| 36 | DOWN_THRESHOLD_H = float(os.environ.get("SOURCE_DOWN_THRESHOLD_H", "48")) |
| 37 | |
| 38 | SOURCE_NODE_MAP: dict[str, list[str]] = { |
| 39 | "proxmox": ["pve-node"], |
| 40 | "white-pc": ["desktop-qi823gr"], |
| 41 | "3060": ["desktop-f8o1v2s"], |
| 42 | "browser-history": ["desktop-f8o1v2s-browser"], |
| 43 | "agents": ["ct215-shell-history", "ct241-shell-history"], |
| 44 | } |
| 45 | |
| 46 | |
| 47 | async def _fetch_bridge_sources() -> dict: |
| 48 | async with httpx.AsyncClient(timeout=3.0) as client: |
| 49 | resp = await client.get(f"{RAG_BRIDGE_URL}/sources") |
| 50 | resp.raise_for_status() |
| 51 | return resp.json() |
| 52 | |
| 53 | |
| 54 | async def _fetch_deck_agents() -> int | None: |
| 55 | try: |
| 56 | async with httpx.AsyncClient(timeout=2.5) as client: |
| 57 | resp = await client.get(f"{DECK_URL}/health") |
| 58 | resp.raise_for_status() |
| 59 | return int(resp.json().get("active_agents", 0)) |
| 60 | except Exception: |
| 61 | return None |
| 62 | |
| 63 | |
| 64 | def _parse_iso(val: str | None) -> datetime | None: |
| 65 | if not val: |
| 66 | return None |
| 67 | try: |
| 68 | return datetime.fromisoformat(str(val).replace("Z", "+00:00")) |
| 69 | except Exception: |
| 70 | return None |
| 71 | |
| 72 | |
| 73 | async def source_statuses() -> list[EngineStatus]: |
| 74 | """Return one EngineStatus per organism data-source node. |
| 75 | |
| 76 | Same schema as the engine entries (engine=node-key, status up/down, |
| 77 | last_ingest, staleness_h). Graceful: if the bridge is down, every node is |
| 78 | emitted status:"down" so the organism withers rather than vanishing. |
| 79 | """ |
| 80 | try: |
| 81 | payload = await _fetch_bridge_sources() |
| 82 | by_id = { |
| 83 | s["source_id"]: s |
| 84 | for s in payload.get("sources", []) |
| 85 | if isinstance(s, dict) and s.get("source_id") |
| 86 | } |
| 87 | except Exception: |
| 88 | by_id = {} |
| 89 | |
| 90 | out: list[EngineStatus] = [] |
| 91 | for node_key, source_ids in SOURCE_NODE_MAP.items(): |
| 92 | best: dict | None = None |
| 93 | for sid in source_ids: |
| 94 | s = by_id.get(sid) |
| 95 | if not s: |
| 96 | continue |
| 97 | sh = s.get("staleness_h") |
| 98 | if best is None or (sh is not None and ( |
| 99 | best.get("staleness_h") is None or sh < best["staleness_h"] |
| 100 | )): |
| 101 | best = s |
| 102 | |
| 103 | if best is None: |
| 104 | out.append(EngineStatus(engine=node_key, status="down")) |
| 105 | continue |
| 106 | |
| 107 | staleness_h = best.get("staleness_h") |
| 108 | status = "up" |
| 109 | if staleness_h is None or staleness_h > DOWN_THRESHOLD_H: |
| 110 | status = "down" |
| 111 | out.append(EngineStatus( |
| 112 | engine=node_key, |
| 113 | status=status, |
| 114 | last_ingest=_parse_iso(best.get("last_ingest")), |
| 115 | staleness_h=staleness_h, |
| 116 | )) |
| 117 | |
| 118 | active = await _fetch_deck_agents() |
| 119 | if active is not None: |
| 120 | for st in out: |
| 121 | if st.engine == "agents": |
| 122 | st.ms = active |
| 123 | return out |