- Added functionality to persist the current dev_mode in the local state.

- Implemented logic to reset the dataset name when switching from dev_mode to non-dev_mode.
-  tests to cover scenarios involving transitions between dev_mode and non-dev_mode with explicit dataset names.

Author: alkaline-0

dlt/common/pipeline.py

@@ -479,6 +479,8 @@ class TPipelineLocalState(TypedDict, total=False):
     """Run dir when pipeline was instantiated for a first time, defaults to cwd on OSS run context"""
     last_run_context: Optional[TLastRunContext]
     """Context from the last successful pipeline run or sync"""
+    _last_dev_mode: bool
+    """Indicates whether previous run used dev_mode; used to reset state on dev->non-dev toggle"""
 
 
 class TPipelineState(TVersionedState, total=False):

dlt/pipeline/pipeline.py

@@ -7,6 +7,7 @@
     Any,
     Callable,
     ClassVar,
+    Dict,
     List,
     Iterator,
     Optional,
@@ -849,6 +850,10 @@ def _sync_destination(
                         )
             # write the state back
             self._props_to_state(state)
+            # persist current dev_mode in local state so future inits can detect toggles
+            if state.get("_local") is None:
+                state.setdefault("_local", {})
+            state["_local"]["_last_dev_mode"] = self.dev_mode
             # verify state
             if state_default_schema_name := state.get("default_schema_name"):
                 # at least empty list is present
@@ -1582,6 +1587,11 @@ def _get_state(self) -> TPipelineState:
             _local = migrated_state["_local"]
             if "initial_cwd" not in _local:
                 _local["initial_cwd"] = os.path.abspath(dlt.current.run_context().local_dir)
+            # if previous run used dev_mode=True and current instance uses dev_mode=False,
+            # reset pipeline by returning an empty state
+            _last_dev_mode = _local.get("_last_dev_mode")
+            if _last_dev_mode and not self.dev_mode:
+                return default_pipeline_state()
             return migrated_state
         except FileNotFoundError:
             # do not set the state hash, this will happen on first merge
@@ -1725,7 +1735,7 @@ def _props_to_state(self, state: TPipelineState) -> TPipelineState:
                 state[prop] = getattr(self, prop)  # type: ignore
         for prop in Pipeline.LOCAL_STATE_PROPS:
             if not prop.startswith("_"):
-                state["_local"][prop] = getattr(self, prop)  # type: ignore
+                state["_local"][prop] = getattr(self, prop)
         if self._destination:
             state["destination_type"] = self._destination.destination_type
             state["destination_name"] = self._destination.configured_name
@@ -1771,7 +1781,12 @@ def _bump_version_and_extract_state(
 
         Storage will be created on demand. In that case the extracted package will be immediately committed.
         """
-        _, hash_, _ = bump_pipeline_state_version_if_modified(self._props_to_state(state))
+        # Write pipeline props into state and persist the current dev_mode value in local state
+        self._props_to_state(state)
+        if state.get("_local") is None:
+            state.setdefault("_local", {})
+        state["_local"]["_last_dev_mode"] = self.dev_mode
+        _, hash_, _ = bump_pipeline_state_version_if_modified(state)
         should_extract = hash_ != state["_local"].get("_last_extracted_hash")
         if should_extract and extract_state:
             extract_ = extract or Extract(self._schema_storage, self._normalize_storage_config())

tests/pipeline/test_pipeline.py

@@ -297,6 +297,26 @@ def test_run_dev_mode_default_dataset() -> None:
     assert p.dataset_name and p.dataset_name.endswith(p._pipeline_instance_id)
 
 
+def test_dev_mode_then_non_dev_resets_dataset_name() -> None:
+    p_dev = dlt.pipeline(dev_mode=True, destination="filesystem")
+    assert p_dev.dataset_name.endswith(p_dev._pipeline_instance_id)
+    p_non_dev = dlt.pipeline(destination="filesystem")
+    assert not p_non_dev.dataset_name.endswith(p_dev._pipeline_instance_id)
+    assert p_non_dev.dataset_name.endswith("_dataset")
+
+
+def test_dev_mode_then_non_dev_with_explicit_dataset_name() -> None:
+    base_ds = "my_custom_dataset_name"
+    # In dev mode, explicit dataset_name should be suffixed with instance id
+    p_dev = dlt.pipeline(dev_mode=True, destination="filesystem", dataset_name=base_ds)
+    assert p_dev.dataset_name.startswith(base_ds)
+    assert p_dev.dataset_name != base_ds
+    assert p_dev.dataset_name.endswith(p_dev._pipeline_instance_id)
+    # In non-dev mode, explicit dataset_name should be used as-is (no suffix)
+    p_non_dev = dlt.pipeline(destination="filesystem", dataset_name=base_ds)
+    assert p_non_dev.dataset_name == base_ds
+
+
 def test_run_dev_mode_default_dataset_layout(environment) -> None:
     # Set dataset_name_layout to "bobby_%s"
     dataset_name_layout = "bobby_%s"