Merge branch 'feature/open-ai-agents' of https://github.com/UiPath/uipath-llamaindex-python into feature/open-ai-agents

Author: cosmyo

packages/uipath-openai-agents/src/uipath_openai_agents/runtime/factory.py

@@ -72,7 +72,7 @@ def _remove_file_with_retry(self, path: str, max_attempts: int = 5) -> None:
         """Remove file with retry logic for Windows file locking.
 
         OpenAI SDK uses sync sqlite3 which doesn't immediately release file locks
-        on Windows. This retry mechanism gives the OS time to release the lock.
+        on Windows. This method ensures proper cleanup before deletion.
 
         Args:
             path: Path to file to remove
@@ -81,8 +81,31 @@ def _remove_file_with_retry(self, path: str, max_attempts: int = 5) -> None:
         Raises:
             OSError: If file cannot be removed after all retries
         """
+        import gc
+        import sqlite3
         import time
 
+        # Before attempting deletion, ensure the SQLite database is properly closed
+        # by performing a WAL checkpoint to release all file locks
+        try:
+            conn = sqlite3.connect(path, timeout=1.0)
+            try:
+                # Commit any pending transactions
+                conn.commit()
+                # Force WAL checkpoint to release shared memory files
+                conn.execute("PRAGMA wal_checkpoint(TRUNCATE)")
+                conn.commit()
+            finally:
+                conn.close()
+                # Force garbage collection to release file handles
+                gc.collect()
+                # Give OS time to release file locks (especially on Windows)
+                time.sleep(0.1)
+        except Exception:
+            # If checkpoint fails, still try to delete
+            pass
+
+        # Now attempt deletion with retry logic
         for attempt in range(max_attempts):
             try:
                 os.remove(path)

packages/uipath-openai-agents/src/uipath_openai_agents/runtime/runtime.py

@@ -207,7 +207,6 @@ async def _run_agent(
             Runtime events if stream_events=True, then final result
         """
         agent_input = self._prepare_agent_input(input)
-        is_resuming = bool(options and options.resume)
 
         # Create session for state persistence (local to this run)
         # SQLiteSession automatically loads existing data from the database when created
@@ -232,20 +231,12 @@ async def _run_agent(
                 )
                 yield self._create_success_result(result.final_output)
 
-        except Exception:
-            # Clean up session on error
-            if session and self.storage_path and not is_resuming:
-                # Delete incomplete session
-                try:
-                    import os
-
-                    if os.path.exists(self.storage_path):
-                        os.remove(self.storage_path)
-                except Exception:
-                    pass  # Best effort cleanup
-            raise
         finally:
             # Always close session after run completes with proper WAL checkpoint
+            # Note: We don't delete the database file on error because:
+            # 1. Multiple sessions (different session_ids) coexist in the same file
+            # 2. Deleting the file would delete ALL sessions, not just the failed one
+            # 3. File deletion is handled at the factory level when starting new executions
             if session:
                 self._close_session_with_checkpoint(session)
 
@@ -497,6 +488,10 @@ def _close_session_with_checkpoint(self, session: SQLiteSession) -> None:
         Args:
             session: The SQLiteSession to close
         """
+        import gc
+        import platform
+        import time
+
         try:
             # Get the underlying connection
             conn = session._get_connection()
@@ -525,6 +520,16 @@ def _close_session_with_checkpoint(self, session: SQLiteSession) -> None:
             except Exception:
                 pass  # Best effort
 
+            # Force garbage collection to close any open file handles
+            # This is especially important on Windows where file handles
+            # may remain open even after explicit close()
+            gc.collect()
+
+            # On Windows, give the OS time to release file locks
+            # Virus scanners and backup agents often interfere
+            if platform.system() == "Windows":
+                time.sleep(0.1)
+
     async def dispose(self) -> None:
         """Cleanup runtime resources."""
         # Sessions are closed immediately after each run in _run_agent()