update match_documents

app/supabase_db.py

@@ -3,6 +3,8 @@ from postgrest.types import CountMethod
 from supabase.client import create_client, Client
 from loguru import logger
 import re
+import time
+import httpx
 
 from .utils import timing_decorator_sync
 from .constants import VEHICLE_KEYWORD_TO_COLUMN, VIETNAMESE_STOP_WORDS, VIETNAMESE_STOP_PHRASES
@@ -44,7 +46,7 @@ class SupabaseClient:
             return None
 
     @timing_decorator_sync
-    def match_documents(self, embedding: List[float], match_count: Optional[int] = None, vehicle_keywords: Optional[List[str]] = None, user_question: str = '', min_rank_threshold: float = 0.001, rrf_k: int = 60):
+    def match_documents(self, embedding: List[float], match_count: Optional[int] = None, vehicle_keywords: Optional[List[str]] = None, user_question: str = '', keyword_threshold: float = 0.01, vector_threshold: float = 0.3, rrf_k: int = 60):
         """
         Truy vấn vector similarity search qua RPC match_documents.
         Input: embedding (list[float]), match_count (int), vehicle_keywords (list[str] hoặc None)
@@ -74,30 +76,42 @@ class SupabaseClient:
         # logger.info(f"[DEBUG][RPC]: embedding: {embedding[:5]}...{embedding[-5:]}")
         logger.info(f"[DEBUG][RPC]: embedding: {embedding}")
 
-        try:
-            payload = {
-                'query_text': or_query_tsquery,
-                'query_embedding': embedding,
-                'match_count': match_count,
-                'min_rank_threshold': min_rank_threshold,
-                'vehicle_filters': None,
-                'rrf_k': rrf_k
-            }
-            if vehicle_keywords:
-                vehicle_columns = [VEHICLE_KEYWORD_TO_COLUMN[k] for k in vehicle_keywords if k in VEHICLE_KEYWORD_TO_COLUMN]
-                if vehicle_columns:
-                    payload['vehicle_filters'] = vehicle_columns
-            response = self.client.rpc(
-                'match_documents',
-                payload
-            ).execute()
+        payload = {
+            'query_text': or_query_tsquery,
+            'query_embedding': embedding,
+            'match_count': match_count,
+            'keyword_threshold': keyword_threshold,
+            'vector_threshold': vector_threshold,
+            'vehicle_filters': None,
+            'rrf_k': rrf_k
+        }
+        if vehicle_keywords:
+            vehicle_columns = [VEHICLE_KEYWORD_TO_COLUMN[k] for k in vehicle_keywords if k in VEHICLE_KEYWORD_TO_COLUMN]
+            if vehicle_columns:
+                payload['vehicle_filters'] = vehicle_columns
 
-            if response.data:
-                return response.data
-            return []
-        except Exception as e:
-            logger.error(f"Error matching documents: {e}")
-            return []
+        max_retries = 3
+        for attempt in range(max_retries):
+            try:
+                response = self.client.rpc(
+                    'match_documents',
+                    payload
+                ).execute()
+
+                if response.data:
+                    return response.data
+                return []
+            except httpx.TimeoutException:
+                logger.warning(f"Supabase RPC 'match_documents' timeout on attempt {attempt + 1}/{max_retries}. Retrying...")
+                if attempt == max_retries - 1:
+                    logger.error(f"Supabase RPC failed after {max_retries} attempts due to timeout.")
+                    return []
+                time.sleep(1 * (2 ** attempt))  # Exponential backoff: 1s, 2s, 4s
+            except Exception as e:
+                logger.error(f"Error matching documents: {e}")
+                return []
+        
+        return [] # Fallback in case loop finishes without returning
 
     @timing_decorator_sync
     def store_embedding(self, text: str, embedding: List[float], metadata: Dict[str, Any]):