llm-api/plan_router.py aktualisiert

llm-api/plan_router.py

@@ -1,10 +1,12 @@
 # -*- coding: utf-8 -*-
 """
-plan_router.py – v0.10.0 (WP-15)
+plan_router.py – v0.11.0 (WP-15)
 
 Minimal-CRUD für Plan-Templates & Pläne (POST/GET) + Idempotenz via Fingerprint.
-Erweiterung ggü. v0.9.0: optionale Section-Felder ideal/supplement + materialisierte
-Facettenfelder für Qdrant-Indizes; robustere Payload→Model-Konvertierung.
+Erweiterungen:
+- optionale Section-Felder ideal/supplement + materialisierte Facettenfelder
+- Referenz-Validierung: template_id (pflicht, wenn gesetzt)
+- Optionaler Strict-Mode: PLAN_STRICT_EXERCISES prüft exercise_external_id
 """
 from fastapi import APIRouter, HTTPException
 from pydantic import BaseModel, Field
@@ -26,6 +28,7 @@ router = APIRouter(tags=["plans"])
 PLAN_COLLECTION = os.getenv("PLAN_COLLECTION") or os.getenv("QDRANT_COLLECTION_PLANS", "plans")
 PLAN_TEMPLATE_COLLECTION = os.getenv("PLAN_TEMPLATE_COLLECTION", "plan_templates")
 PLAN_SESSION_COLLECTION = os.getenv("PLAN_SESSION_COLLECTION", "plan_sessions")
+EXERCISE_COLLECTION = os.getenv("EXERCISE_COLLECTION", "exercises")
 
 # -----------------
 # Modelle
@@ -34,8 +37,8 @@ class TemplateSection(BaseModel):
     name: str
     target_minutes: int
     must_keywords: List[str] = []
-    ideal_keywords: List[str] = []          # NEU
-    supplement_keywords: List[str] = []     # NEU
+    ideal_keywords: List[str] = []          # optional
+    supplement_keywords: List[str] = []     # optional
     forbid_keywords: List[str] = []
     capability_targets: Dict[str, int] = {}
 
@@ -156,6 +159,16 @@ def _as_model(model_cls, payload: Dict[str, Any]):
     data = {k: payload[k] for k in payload.keys() if k in allowed}
     return model_cls(**data)
 
+
+def _truthy(val: Optional[str]) -> bool:
+    return str(val or "").strip().lower() in {"1", "true", "yes", "on"}
+
+
+def _exists_in_collection(collection: str, key: str, value: Any) -> bool:
+    flt = Filter(must=[FieldCondition(key=key, match=MatchValue(value=value))])
+    pts, _ = qdrant.scroll(collection_name=collection, scroll_filter=flt, limit=1, with_payload=False)
+    return bool(pts)
+
 # -----------------
 # Endpoints
 # -----------------
@@ -168,8 +181,8 @@ def create_plan_template(t: PlanTemplate):
     sections = payload.get("sections", []) or []
     for s in sections:
         s["must_keywords"] = _norm_list(s.get("must_keywords") or [])
-        s["ideal_keywords"] = _norm_list(s.get("ideal_keywords") or [])            # NEU
-        s["supplement_keywords"] = _norm_list(s.get("supplement_keywords") or [])  # NEU
+        s["ideal_keywords"] = _norm_list(s.get("ideal_keywords") or [])
+        s["supplement_keywords"] = _norm_list(s.get("supplement_keywords") or [])
         s["forbid_keywords"] = _norm_list(s.get("forbid_keywords") or [])
 
     # Materialisierte Facettenfelder (stabile KEYWORD-Indizes in Qdrant)
@@ -196,19 +209,38 @@ def get_plan_template(tpl_id: str):
 @router.post("/plan", response_model=Plan)
 def create_plan(p: Plan):
     _ensure_collection(PLAN_COLLECTION)
-    # Fingerprint
+
+    # 1) Template-Referenz prüfen (falls gesetzt)
+    if p.template_id:
+        if not _exists_in_collection(PLAN_TEMPLATE_COLLECTION, "id", p.template_id):
+            raise HTTPException(status_code=422, detail=f"Unknown template_id: {p.template_id}")
+
+    # 2) Optional: Strict-Mode – Exercises prüfen
+    if _truthy(os.getenv("PLAN_STRICT_EXERCISES")):
+        missing: List[str] = []
+        for sec in p.sections or []:
+            for it in sec.items or []:
+                exid = (it.exercise_external_id or "").strip()
+                if exid and not _exists_in_collection(EXERCISE_COLLECTION, "external_id", exid):
+                    missing.append(exid)
+        if missing:
+            raise HTTPException(status_code=422, detail={
+                "error": "unknown exercise_external_id",
+                "missing": sorted(set(missing))
+            })
+
+    # 3) Fingerprint
     fp = _fingerprint_for_plan(p)
     p.fingerprint = p.fingerprint or fp
 
-    # Idempotenz
+    # 4) Idempotenz
     existing = _get_by_field(PLAN_COLLECTION, "fingerprint", p.fingerprint)
     if existing:
         return _as_model(Plan, existing)
 
-    # Normalisieren
+    # 5) Normalisieren & upsert
     p.goals = _norm_list(p.goals)
     payload = p.model_dump()
-    # ISO8601 sicherstellen
     if isinstance(payload.get("created_at"), datetime):
         payload["created_at"] = payload["created_at"].astimezone(timezone.utc).isoformat()
 
@@ -223,7 +255,6 @@ def get_plan(plan_id: str):
     found = _get_by_field(PLAN_COLLECTION, "id", plan_id)
     if not found:
         raise HTTPException(status_code=404, detail="not found")
-    # created_at zurück in datetime (ISO)
     if isinstance(found.get("created_at"), str):
         try:
             found["created_at"] = datetime.fromisoformat(found["created_at"])