fix(api): align kb extractor ollama model

apps/api/src/services/knowledge_extractor_service.py

@@ -4,7 +4,7 @@ Knowledge Extractor Service — KB Phase 2-A
 Incident resolved 後自動萃取 KB 草稿。
 
 設計原則:
-- 使用 Ollama llama3.2:3b，依全域順序 GCP-A → GCP-B → 111 嘗試
+- 使用 `settings.OLLAMA_TOOL_MODEL`，依全域順序 GCP-A → GCP-B → 111 嘗試
 - fire-and-forget：失敗不影響 resolve 主流程
 - logger.exception 保留完整 Stack Trace 供 Prompt 調優
 
@@ -17,10 +17,19 @@ logger = structlog.get_logger(__name__)
 
 # 2026-05-19 Codex: 統帥校正，全 Ollama workload 固定 GCP-A → GCP-B → 111。
 def _get_ollama_endpoints():
-    from src.services.ollama_endpoint_resolver import resolve_ollama_order
+    from src.services.ollama_endpoint_circuit_breaker import (
+        resolve_ollama_order_with_cooldown,
+    )
+
+    return resolve_ollama_order_with_cooldown("hermes")
+
+
+def _get_extract_model() -> str:
+    from src.core.config import settings
+
+    return str(getattr(settings, "OLLAMA_TOOL_MODEL", "hermes3:latest") or "hermes3:latest")
+
 
-    return resolve_ollama_order("deep_rca")
-_EXTRACT_MODEL = "llama3.2:3b"
 _EXTRACT_TIMEOUT = 30.0  # 秒，容忍慢速
 
 # Linear / Nothing.tech 風格的 SRE KB Prompt
@@ -72,7 +81,7 @@ class KnowledgeExtractorService:
     """
     Incident → KB 草稿自動萃取器
 
-    使用 Ollama llama3.2:3b 本地推理，產生 Markdown 格式的 SRE 知識條目。
+    使用目前配置的 Ollama tool model 產生 Markdown 格式的 SRE 知識條目。
     """
 
     async def extract_from_incident(self, incident) -> bool:
@@ -103,11 +112,12 @@ class KnowledgeExtractorService:
 
             # 2. 呼叫 Ollama（直接 HTTP，不走 AIRouter 避免路由邏輯開銷）
             markdown_content = await self._call_ollama(prompt)
+            model = _get_extract_model()
             if not markdown_content:
                 logger.warning(
                     "kb_extract_empty_response",
                     incident_id=incident.incident_id,
-                    model=_EXTRACT_MODEL,
+                    model=model,
                 )
                 return False
 
@@ -142,7 +152,7 @@ class KnowledgeExtractorService:
                 incident_id=incident.incident_id,
                 title=title,
                 category=category,
-                model=_EXTRACT_MODEL,
+                model=model,
             )
             return True
 
@@ -165,6 +175,7 @@ class KnowledgeExtractorService:
         import httpx
 
         endpoints = _get_ollama_endpoints()
+        model = _get_extract_model()
         async with httpx.AsyncClient(timeout=_EXTRACT_TIMEOUT) as client:
             for endpoint in endpoints:
                 if not endpoint.url:
@@ -173,7 +184,7 @@ class KnowledgeExtractorService:
                     r = await client.post(
                         f"{endpoint.url}/api/generate",
                         json={
-                            "model": _EXTRACT_MODEL,
+                            "model": model,
                             "prompt": prompt,
                             "stream": False,
                             "options": {
@@ -188,15 +199,25 @@ class KnowledgeExtractorService:
                     if text:
                         logger.info(
                             "kb_ollama_call_success",
-                            model=_EXTRACT_MODEL,
+                            model=model,
                             provider=endpoint.provider_name,
                             base=endpoint.url,
                         )
+                        from src.services.ollama_endpoint_circuit_breaker import (
+                            record_ollama_endpoint_success,
+                        )
+
+                        record_ollama_endpoint_success(endpoint.url)
                         return text
                 except Exception as e:
+                    from src.services.ollama_endpoint_circuit_breaker import (
+                        record_ollama_endpoint_failure,
+                    )
+
+                    record_ollama_endpoint_failure(endpoint.url)
                     logger.warning(
                         "kb_ollama_call_failed",
-                        model=_EXTRACT_MODEL,
+                        model=model,
                         provider=endpoint.provider_name,
                         base=endpoint.url,
                         error=str(e),
@@ -204,7 +225,7 @@ class KnowledgeExtractorService:
 
         logger.error(
             "kb_ollama_all_endpoints_failed",
-            model=_EXTRACT_MODEL,
+            model=model,
             attempted=[endpoint.provider_name for endpoint in endpoints],
         )
         return None