fix(execute_class): address PR review feedback

- Replace misleading carried-over comment with accurate description
- Add inline comments explaining double-checked locking pattern
- Add failure-path test: deploy exception releases lock, allows retry

Author: deanq

src/runpod_flash/execute_class.py

@@ -225,23 +225,22 @@ def __init__(self, *args, **kwargs):
             )
 
         async def _ensure_initialized(self):
-            """Ensure the remote instance is created."""
+            """Ensure the remote instance is created exactly once, even under concurrent calls."""
+            # Fast path: already initialized, no lock needed.
             if self._initialized:
                 return
 
+            # Slow path: acquire lock and re-check to prevent double deployment
+            # when multiple coroutines race past the fast-path check.
             async with self._init_lock:
                 if self._initialized:
                     return
 
-                # Get remote resource
                 resource_manager = ResourceManager()
                 remote_resource = await resource_manager.get_or_deploy_resource(
                     self._resource_config
                 )
                 self._stub = stub_resource(remote_resource, **self._extra)
-
-                # Create the remote instance by calling a method (which will trigger instance creation)
-                # We'll do this on first method call
                 self._initialized = True
 
         def __getattr__(self, name):

tests/bug_probes/test_class_execution.py

@@ -111,3 +111,35 @@ async def test_second_call_skips_deploy(self, wrapper_instance):
 
             await wrapper_instance._ensure_initialized()
             mock_rm.get_or_deploy_resource.assert_awaited_once()
+
+    @pytest.mark.asyncio
+    async def test_deploy_failure_releases_lock_and_allows_retry(
+        self, wrapper_instance
+    ):
+        """If deploy fails, the lock must be released and a subsequent call must retry."""
+        call_count = 0
+
+        async def failing_then_succeeding_deploy(config):
+            nonlocal call_count
+            call_count += 1
+            if call_count == 1:
+                raise ConnectionError("transient failure")
+            return MagicMock()
+
+        with (
+            patch("runpod_flash.execute_class.ResourceManager") as mock_rm_cls,
+            patch("runpod_flash.execute_class.stub_resource", return_value=MagicMock()),
+        ):
+            mock_rm = MagicMock()
+            mock_rm.get_or_deploy_resource = failing_then_succeeding_deploy
+            mock_rm_cls.return_value = mock_rm
+
+            with pytest.raises(ConnectionError, match="transient failure"):
+                await wrapper_instance._ensure_initialized()
+
+            assert not wrapper_instance._initialized
+
+            # Retry should succeed
+            await wrapper_instance._ensure_initialized()
+            assert wrapper_instance._initialized
+            assert call_count == 2