修复使用pgvector时get_doc_by_ids方法报错的bug及知识库文件删除后向量仍然存在的bug (#3407)

* 修复使用pgvector无法查询到document

* 修复使用pg/es向量库无法知识库删除文档后向量库并未删除对应记录的bug

server/knowledge_base/kb_service/base.py

@@ -222,6 +222,20 @@ class KBService(ABC):
                 pass
         return docs
 
+    def get_relative_source_path(self,filepath: str):
+      '''
+      将文件路径转化为相对路径，保证查询时一致
+      '''
+      relative_path = filepath
+      if os.path.isabs(relative_path):
+        try:
+          relative_path = Path(filepath).relative_to(self.doc_path)
+        except Exception as e:
+          print(f"cannot convert absolute path ({source}) to relative path. error is : {e}")
+
+      relative_path = str(relative_path.as_posix().strip("/"))
+      return relative_path
+
     @abstractmethod
     def do_create_kb(self):
         """

server/knowledge_base/kb_service/es_kb_service.py

@@ -184,7 +184,7 @@ class ESKBService(KBService):
             query = {
                 "query": {
                     "term": {
-                        "metadata.source.keyword": kb_file.filepath
+                        "metadata.source.keyword": self.get_relative_source_path(kb_file.filepath)
                     }
                 }
             }

server/knowledge_base/kb_service/pg_kb_service.py

@@ -28,9 +28,9 @@ class PGKBService(KBService):
 
     def get_doc_by_ids(self, ids: List[str]) -> List[Document]:
         with Session(PGKBService.engine) as session:
-            stmt = text("SELECT document, cmetadata FROM langchain_pg_embedding WHERE collection_id in :ids")
+            stmt = text("SELECT document, cmetadata FROM langchain_pg_embedding WHERE custom_id = ANY(:ids)")
             results = [Document(page_content=row[0], metadata=row[1]) for row in
-                       session.execute(stmt, {'ids': ids}).fetchall()]
+                      session.execute(stmt, {'ids': ids}).fetchall()]
             return results
     def del_doc_by_ids(self, ids: List[str]) -> bool:
         return super().del_doc_by_ids(ids)
@@ -71,11 +71,10 @@ class PGKBService(KBService):
 
     def do_delete_doc(self, kb_file: KnowledgeFile, **kwargs):
         with Session(PGKBService.engine) as session:
-            filepath = kb_file.filepath.replace('\\', '\\\\')
             session.execute(
                 text(
                     ''' DELETE FROM langchain_pg_embedding WHERE cmetadata::jsonb @> '{"source": "filepath"}'::jsonb;'''.replace(
-                        "filepath", filepath)))
+                        "filepath", self.get_relative_source_path(kb_file.filepath))))
             session.commit()
 
     def do_clear_vs(self):