flwr-nightly 1.8.0.dev20240304__py3-none-any.whl → 1.8.0.dev20240306__py3-none-any.whl

flwr/simulation/ray_transport/ray_client_proxy.py

@@ -24,12 +24,7 @@ from flwr.client import ClientFn
 from flwr.client.client_app import ClientApp
 from flwr.client.node_state import NodeState
 from flwr.common import Message, Metadata, RecordSet
-from flwr.common.constant import (
-    MESSAGE_TYPE_EVALUATE,
-    MESSAGE_TYPE_FIT,
-    MESSAGE_TYPE_GET_PARAMETERS,
-    MESSAGE_TYPE_GET_PROPERTIES,
-)
+from flwr.common.constant import MessageType, MessageTypeLegacy
 from flwr.common.logger import log
 from flwr.common.recordset_compat import (
     evaluateins_to_recordset,
@@ -126,7 +121,7 @@ class RayActorClientProxy(ClientProxy):
         recordset = getpropertiesins_to_recordset(ins)
         message = self._wrap_recordset_in_message(
             recordset,
-            message_type=MESSAGE_TYPE_GET_PROPERTIES,
+            message_type=MessageTypeLegacy.GET_PROPERTIES,
             timeout=timeout,
             group_id=group_id,
         )
@@ -145,7 +140,7 @@ class RayActorClientProxy(ClientProxy):
         recordset = getparametersins_to_recordset(ins)
         message = self._wrap_recordset_in_message(
             recordset,
-            message_type=MESSAGE_TYPE_GET_PARAMETERS,
+            message_type=MessageTypeLegacy.GET_PARAMETERS,
             timeout=timeout,
             group_id=group_id,
         )
@@ -163,7 +158,7 @@ class RayActorClientProxy(ClientProxy):
         )  # This must stay TRUE since ins are in-memory
         message = self._wrap_recordset_in_message(
             recordset,
-            message_type=MESSAGE_TYPE_FIT,
+            message_type=MessageType.TRAIN,
             timeout=timeout,
             group_id=group_id,
         )
@@ -181,7 +176,7 @@ class RayActorClientProxy(ClientProxy):
         )  # This must stay TRUE since ins are in-memory
         message = self._wrap_recordset_in_message(
             recordset,
-            message_type=MESSAGE_TYPE_EVALUATE,
+            message_type=MessageType.EVALUATE,
             timeout=timeout,
             group_id=group_id,
         )

flwr/simulation/run_simulation.py

@@ -17,39 +17,153 @@
 import argparse
 import asyncio
 import json
+import logging
 import threading
 import traceback
-from logging import ERROR, INFO, WARNING
+from logging import DEBUG, ERROR, INFO, WARNING
 from time import sleep
-from typing import Any, Callable
+from typing import Dict, Optional
 
 import grpc
 
+from flwr.client import ClientApp
 from flwr.common import EventType, event, log
+from flwr.common.typing import ConfigsRecordValues
 from flwr.server.driver.driver import Driver
 from flwr.server.run_serverapp import run
+from flwr.server.server_app import ServerApp
 from flwr.server.superlink.driver.driver_grpc import run_driver_api_grpc
 from flwr.server.superlink.fleet import vce
 from flwr.server.superlink.state import StateFactory
-from flwr.simulation.ray_transport.utils import enable_tf_gpu_growth
+from flwr.simulation.ray_transport.utils import (
+    enable_tf_gpu_growth as enable_gpu_growth,
+)
 
 
+# Entry point from CLI
+def run_simulation_from_cli() -> None:
+    """Run Simulation Engine from the CLI."""
+    args = _parse_args_run_simulation().parse_args()
+
+    # Load JSON config
+    backend_config_dict = json.loads(args.backend_config)
+
+    _run_simulation(
+        server_app_attr=args.server_app,
+        client_app_attr=args.client_app,
+        num_supernodes=args.num_supernodes,
+        backend_name=args.backend,
+        backend_config=backend_config_dict,
+        app_dir=args.app_dir,
+        driver_api_address=args.driver_api_address,
+        enable_tf_gpu_growth=args.enable_tf_gpu_growth,
+        verbose_logging=args.verbose,
+    )
+
+
+# Entry point from Python session (script or notebook)
+# pylint: disable=too-many-arguments
+def run_simulation(
+    server_app: ServerApp,
+    client_app: ClientApp,
+    num_supernodes: int,
+    backend_name: str = "ray",
+    backend_config: Optional[Dict[str, ConfigsRecordValues]] = None,
+    enable_tf_gpu_growth: bool = False,
+    verbose_logging: bool = False,
+) -> None:
+    r"""Run a Flower App using the Simulation Engine.
+
+    Parameters
+    ----------
+    server_app : ServerApp
+        The `ServerApp` to be executed. It will send messages to different `ClientApp`
+        instances running on different (virtual) SuperNodes.
+
+    client_app : ClientApp
+        The `ClientApp` to be executed by each of the SuperNodes. It will receive
+        messages sent by the `ServerApp`.
+
+    num_supernodes : int
+        Number of nodes that run a ClientApp. They can be sampled by a
+        Driver in the ServerApp and receive a Message describing what the ClientApp
+        should perform.
+
+    backend_name : str (default: ray)
+        A simulation backend that runs `ClientApp`s.
+
+    backend_config : Optional[Dict[str, ConfigsRecordValues]]
+        'A dictionary, e.g {"<keyA>": <value>, "<keyB>": <value>} to configure a
+        backend. Values supported in <value> are those included by
+        `flwr.common.typing.ConfigsRecordValues`.
+
+    enable_tf_gpu_growth : bool (default: False)
+        A boolean to indicate whether to enable GPU growth on the main thread. This is
+        desirable if you make use of a TensorFlow model on your `ServerApp` while
+        having your `ClientApp` running on the same GPU. Without enabling this, you
+        might encounter an out-of-memory error because TensorFlow, by default, allocates
+        all GPU memory. Read more about how `tf.config.experimental.set_memory_growth()`
+        works in the TensorFlow documentation: https://www.tensorflow.org/api/stable.
+
+    verbose_logging : bool (default: False)
+        When diabled, only INFO, WARNING and ERROR log messages will be shown. If
+        enabled, DEBUG-level logs will be displayed.
+    """
+    _run_simulation(
+        num_supernodes=num_supernodes,
+        client_app=client_app,
+        server_app=server_app,
+        backend_name=backend_name,
+        backend_config=backend_config,
+        enable_tf_gpu_growth=enable_tf_gpu_growth,
+        verbose_logging=verbose_logging,
+    )
+
+
+# pylint: disable=too-many-arguments
 def run_serverapp_th(
-    server_app_attr: str,
+    server_app_attr: Optional[str],
+    server_app: Optional[ServerApp],
     driver: Driver,
-    server_app_dir: str,
+    app_dir: str,
     f_stop: asyncio.Event,
+    enable_tf_gpu_growth: bool,
     delay_launch: int = 3,
 ) -> threading.Thread:
     """Run SeverApp in a thread."""
+
+    def server_th_with_start_checks(  # type: ignore
+        tf_gpu_growth: bool, stop_event: asyncio.Event, **kwargs
+    ) -> None:
+        """Run SeverApp, after check if GPU memory grouwth has to be set.
+
+        Upon exception, trigger stop event for Simulation Engine.
+        """
+        try:
+            if tf_gpu_growth:
+                log(INFO, "Enabling GPU growth for Tensorflow on the main thread.")
+                enable_gpu_growth()
+
+            # Run ServerApp
+            run(**kwargs)
+        except Exception as ex:  # pylint: disable=broad-exception-caught
+            log(ERROR, "ServerApp thread raised an exception: %s", ex)
+            log(ERROR, traceback.format_exc())
+        finally:
+            log(DEBUG, "ServerApp finished running.")
+            # Upon completion, trigger stop event if one was passed
+            if stop_event is not None:
+                stop_event.set()
+                log(WARNING, "Triggered stop event for Simulation Engine.")
+
     serverapp_th = threading.Thread(
-        target=run,
+        target=server_th_with_start_checks,
+        args=(enable_tf_gpu_growth, f_stop),
         kwargs={
             "server_app_attr": server_app_attr,
+            "loaded_server_app": server_app,
             "driver": driver,
-            "server_app_dir": server_app_dir,
-            "stop_event": f_stop,  # will be set when `run()` finishes
-            # will trigger the shutdown of the Simulation Engine
+            "server_app_dir": app_dir,
         },
     )
     sleep(delay_launch)
@@ -57,52 +171,31 @@ def run_serverapp_th(
     return serverapp_th
 
 
-def get_thread_exception_hook(stop_event: asyncio.Event) -> Callable[[Any], None]:
-    """Return a callback for when the ServerApp thread raises an exception."""
-
-    def execepthook(args: Any) -> None:
-        """Upon exception raised, log exception and trigger stop event."""
-        # log
-        log(
-            ERROR,
-            "The ServerApp thread triggered exception (%s): %s",
-            args.exc_type,
-            args.exc_value,
-        )
-        log(ERROR, traceback.format_exc())
-        # Set stop event
-        stop_event.set()
-        log(WARNING, "Triggered stop event for Simulation Engine.")
-
-    return execepthook
-
-
-def run_simulation() -> None:
-    """Run Simulation Engine."""
-    args = _parse_args_run_simulation().parse_args()
-
-    # Load JSON config
-    backend_config_dict = json.loads(args.backend_config)
-
-    # Enable GPU memory growth (relevant only for TF)
-    if args.enable_tf_gpu_growth:
-        log(INFO, "Enabling GPU growth for Tensorflow on the main thread.")
-        enable_tf_gpu_growth()
-        # Check that Backend config has also enabled using GPU growth
-        use_tf = backend_config_dict.get("tensorflow", False)
-        if not use_tf:
-            log(WARNING, "Enabling GPU growth for your backend.")
-            backend_config_dict["tensorflow"] = True
-
-    # Convert back to JSON stream
-    backend_config = json.dumps(backend_config_dict)
+# pylint: disable=too-many-locals
+def _main_loop(
+    num_supernodes: int,
+    backend_name: str,
+    backend_config_stream: str,
+    driver_api_address: str,
+    app_dir: str,
+    enable_tf_gpu_growth: bool,
+    client_app: Optional[ClientApp] = None,
+    client_app_attr: Optional[str] = None,
+    server_app: Optional[ServerApp] = None,
+    server_app_attr: Optional[str] = None,
+) -> None:
+    """Launch SuperLink with Simulation Engine, then ServerApp on a separate thread.
 
+    Everything runs on the main thread or a separate one, depening on whether the main
+    thread already contains a running Asyncio event loop. This is the case if running
+    the Simulation Engine on a Jupyter/Colab notebook.
+    """
     # Initialize StateFactory
     state_factory = StateFactory(":flwr-in-memory-state:")
 
     # Start Driver API
     driver_server: grpc.Server = run_driver_api_grpc(
-        address=args.driver_api_address,
+        address=driver_api_address,
         state_factory=state_factory,
         certificates=None,
     )
@@ -112,35 +205,39 @@ def run_simulation() -> None:
     try:
         # Initialize Driver
         driver = Driver(
-            driver_service_address=args.driver_api_address,
+            driver_service_address=driver_api_address,
             root_certificates=None,
         )
 
         # Get and run ServerApp thread
-        serverapp_th = run_serverapp_th(args.server_app, driver, args.dir, f_stop)
-        # Setup an exception hook
-        threading.excepthook = get_thread_exception_hook(f_stop)
+        serverapp_th = run_serverapp_th(
+            server_app_attr=server_app_attr,
+            server_app=server_app,
+            driver=driver,
+            app_dir=app_dir,
+            f_stop=f_stop,
+            enable_tf_gpu_growth=enable_tf_gpu_growth,
+        )
 
         # SuperLink with Simulation Engine
         event(EventType.RUN_SUPERLINK_ENTER)
         vce.start_vce(
-            num_supernodes=args.num_supernodes,
-            client_app_module_name=args.client_app,
-            backend_name=args.backend,
-            backend_config_json_stream=backend_config,
-            working_dir=args.dir,
+            num_supernodes=num_supernodes,
+            client_app_attr=client_app_attr,
+            client_app=client_app,
+            backend_name=backend_name,
+            backend_config_json_stream=backend_config_stream,
+            app_dir=app_dir,
             state_factory=state_factory,
             f_stop=f_stop,
         )
 
     except Exception as ex:
-
-        log(ERROR, "An exception occurred: %s", ex)
+        log(ERROR, "An exception occurred !! %s", ex)
         log(ERROR, traceback.format_exc())
         raise RuntimeError("An error was encountered. Ending simulation.") from ex
 
     finally:
-
         # Stop Driver
         driver_server.stop(grace=0)
         del driver
@@ -154,20 +251,148 @@ def run_simulation() -> None:
     log(INFO, "Stopping Simulation Engine now.")
 
 
+# pylint: disable=too-many-arguments,too-many-locals
+def _run_simulation(
+    num_supernodes: int,
+    client_app: Optional[ClientApp] = None,
+    server_app: Optional[ServerApp] = None,
+    backend_name: str = "ray",
+    backend_config: Optional[Dict[str, ConfigsRecordValues]] = None,
+    client_app_attr: Optional[str] = None,
+    server_app_attr: Optional[str] = None,
+    app_dir: str = "",
+    driver_api_address: str = "0.0.0.0:9091",
+    enable_tf_gpu_growth: bool = False,
+    verbose_logging: bool = False,
+) -> None:
+    r"""Launch the Simulation Engine.
+
+    Parameters
+    ----------
+    num_supernodes : int
+        Number of nodes that run a ClientApp. They can be sampled by a
+        Driver in the ServerApp and receive a Message describing what the ClientApp
+        should perform.
+
+    client_app : Optional[ClientApp]
+        The `ClientApp` to be executed by each of the `SuperNodes`. It will receive
+        messages sent by the `ServerApp`.
+
+    server_app : Optional[ServerApp]
+        The `ServerApp` to be executed.
+
+    backend_name : str (default: ray)
+        A simulation backend that runs `ClientApp`s.
+
+    backend_config : Optional[Dict[str, ConfigsRecordValues]]
+        'A dictionary, e.g {"<keyA>":<value>, "<keyB>":<value>} to configure a
+        backend. Values supported in <value> are those included by
+        `flwr.common.typing.ConfigsRecordValues`.
+
+    client_app_attr : str
+        A path to a `ClientApp` module to be loaded: For example: `client:app` or
+        `project.package.module:wrapper.app`."
+
+    server_app_attr : str
+        A path to a `ServerApp` module to be loaded: For example: `server:app` or
+        `project.package.module:wrapper.app`."
+
+    app_dir : str
+        Add specified directory to the PYTHONPATH and load `ClientApp` from there.
+        (Default: current working directory.)
+
+    driver_api_address : str (default: "0.0.0.0:9091")
+        Driver API (gRPC) server address (IPv4, IPv6, or a domain name)
+
+    enable_tf_gpu_growth : bool (default: False)
+        A boolean to indicate whether to enable GPU growth on the main thread. This is
+        desirable if you make use of a TensorFlow model on your `ServerApp` while
+        having your `ClientApp` running on the same GPU. Without enabling this, you
+        might encounter an out-of-memory error becasue TensorFlow by default allocates
+        all GPU memory. Read mor about how `tf.config.experimental.set_memory_growth()`
+        works in the TensorFlow documentation: https://www.tensorflow.org/api/stable.
+
+    verbose_logging : bool (default: False)
+        When diabled, only INFO, WARNING and ERROR log messages will be shown. If
+        enabled, DEBUG-level logs will be displayed.
+    """
+    # Set logging level
+    if not verbose_logging:
+        logger = logging.getLogger("flwr")
+        logger.setLevel(INFO)
+
+    if backend_config is None:
+        backend_config = {}
+
+    if enable_tf_gpu_growth:
+        # Check that Backend config has also enabled using GPU growth
+        use_tf = backend_config.get("tensorflow", False)
+        if not use_tf:
+            log(WARNING, "Enabling GPU growth for your backend.")
+            backend_config["tensorflow"] = True
+
+    # Convert config to original JSON-stream format
+    backend_config_stream = json.dumps(backend_config)
+
+    simulation_engine_th = None
+    args = (
+        num_supernodes,
+        backend_name,
+        backend_config_stream,
+        driver_api_address,
+        app_dir,
+        enable_tf_gpu_growth,
+        client_app,
+        client_app_attr,
+        server_app,
+        server_app_attr,
+    )
+    # Detect if there is an Asyncio event loop already running.
+    # If yes, run everything on a separate thread. In environmnets
+    # like Jupyter/Colab notebooks, there is an event loop present.
+    run_in_thread = False
+    try:
+        _ = (
+            asyncio.get_running_loop()
+        )  # Raises RuntimeError if no event loop is present
+        log(DEBUG, "Asyncio event loop already running.")
+
+        run_in_thread = True
+
+    except RuntimeError:
+        log(DEBUG, "No asyncio event loop runnig")
+
+    finally:
+        if run_in_thread:
+            log(DEBUG, "Starting Simulation Engine on a new thread.")
+            simulation_engine_th = threading.Thread(target=_main_loop, args=args)
+            simulation_engine_th.start()
+            simulation_engine_th.join()
+        else:
+            log(DEBUG, "Starting Simulation Engine on the main thread.")
+            _main_loop(*args)
+
+
 def _parse_args_run_simulation() -> argparse.ArgumentParser:
     """Parse flower-simulation command line arguments."""
     parser = argparse.ArgumentParser(
         description="Start a Flower simulation",
     )
+    parser.add_argument(
+        "--server-app",
+        required=True,
+        help="For example: `server:app` or `project.package.module:wrapper.app`",
+    )
     parser.add_argument(
         "--client-app",
         required=True,
         help="For example: `client:app` or `project.package.module:wrapper.app`",
     )
     parser.add_argument(
-        "--server-app",
+        "--num-supernodes",
+        type=int,
         required=True,
-        help="For example: `server:app` or `project.package.module:wrapper.app`",
+        help="Number of simulated SuperNodes.",
     )
     parser.add_argument(
         "--driver-api-address",
@@ -175,18 +400,20 @@ def _parse_args_run_simulation() -> argparse.ArgumentParser:
         type=str,
         help="For example: `server:app` or `project.package.module:wrapper.app`",
     )
-    parser.add_argument(
-        "--num-supernodes",
-        type=int,
-        required=True,
-        help="Number of simulated SuperNodes.",
-    )
     parser.add_argument(
         "--backend",
         default="ray",
         type=str,
         help="Simulation backend that executes the ClientApp.",
     )
+    parser.add_argument(
+        "--backend-config",
+        type=str,
+        default='{"client_resources": {"num_cpus":2, "num_gpus":0.0}, "tensorflow": 0}',
+        help='A JSON formatted stream, e.g \'{"<keyA>":<value>, "<keyB>":<value>}\' to '
+        "configure a backend. Values supported in <value> are those included by "
+        "`flwr.common.typing.ConfigsRecordValues`. ",
+    )
     parser.add_argument(
         "--enable-tf-gpu-growth",
         action="store_true",
@@ -198,15 +425,13 @@ def _parse_args_run_simulation() -> argparse.ArgumentParser:
         "the TensorFlow documentation: https://www.tensorflow.org/api/stable.",
     )
     parser.add_argument(
-        "--backend-config",
-        type=str,
-        default='{"client_resources": {"num_cpus":2, "num_gpus":0.0}, "tensorflow": 0}',
-        help='A JSON formatted stream, e.g \'{"<keyA>":<value>, "<keyB>":<value>}\' to '
-        "configure a backend. Values supported in <value> are those included by "
-        "`flwr.common.typing.ConfigsRecordValues`. ",
+        "--verbose",
+        action="store_true",
+        help="When unset, only INFO, WARNING and ERROR log messages will be shown. "
+        "If set, DEBUG-level logs will be displayed. ",
     )
     parser.add_argument(
-        "--dir",
+        "--app-dir",
         default="",
         help="Add specified directory to the PYTHONPATH and load"
         "ClientApp and ServerApp from there."

{flwr_nightly-1.8.0.dev20240304.dist-info → flwr_nightly-1.8.0.dev20240306.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: flwr-nightly
-Version: 1.8.0.dev20240304
+Version: 1.8.0.dev20240306
 Summary: Flower: A Friendly Federated Learning Framework
 Home-page: https://flower.ai
 License: Apache-2.0
@@ -42,6 +42,7 @@ Requires-Dist: pydantic (<2.0.0) ; extra == "simulation"
 Requires-Dist: ray (==2.6.3) ; extra == "simulation"
 Requires-Dist: requests (>=2.31.0,<3.0.0) ; extra == "rest"
 Requires-Dist: starlette (>=0.31.0,<0.32.0) ; extra == "rest"
+Requires-Dist: tomli (>=2.0.1,<3.0.0)
 Requires-Dist: typer[all] (>=0.9.0,<0.10.0)
 Requires-Dist: uvicorn[standard] (>=0.23.0,<0.24.0) ; extra == "rest"
 Project-URL: Documentation, https://flower.ai
@@ -182,7 +183,6 @@ Quickstart examples:
 - [Quickstart (JAX)](https://github.com/adap/flower/tree/main/examples/quickstart-jax)
 - [Quickstart (MONAI)](https://github.com/adap/flower/tree/main/examples/quickstart-monai)
 - [Quickstart (scikit-learn)](https://github.com/adap/flower/tree/main/examples/sklearn-logreg-mnist)
-- [Quickstart (XGBoost)](https://github.com/adap/flower/tree/main/examples/xgboost-quickstart)
 - [Quickstart (Android [TFLite])](https://github.com/adap/flower/tree/main/examples/android)
 - [Quickstart (iOS [CoreML])](https://github.com/adap/flower/tree/main/examples/ios)
 - [Quickstart (MLX)](https://github.com/adap/flower/tree/main/examples/quickstart-mlx)
@@ -194,7 +194,8 @@ Other [examples](https://github.com/adap/flower/tree/main/examples):
 - [PyTorch: From Centralized to Federated](https://github.com/adap/flower/tree/main/examples/pytorch-from-centralized-to-federated)
 - [Vertical FL](https://github.com/adap/flower/tree/main/examples/vertical-fl)
 - [Federated Finetuning of OpenAI's Whisper](https://github.com/adap/flower/tree/main/examples/whisper-federated-finetuning)
-- [Comprehensive XGBoost](https://github.com/adap/flower/tree/main/examples/xgboost-comprehensive)
+- [Federated Finetuning of Large Language Model](https://github.com/adap/flower/tree/main/examples/fedllm-finetune)
+- [Federated Finetuning of a Vision Transformer](https://github.com/adap/flower/tree/main/examples/vit-finetune)
 - [Advanced Flower with TensorFlow/Keras](https://github.com/adap/flower/tree/main/examples/advanced-tensorflow)
 - [Advanced Flower with PyTorch](https://github.com/adap/flower/tree/main/examples/advanced-pytorch)
 - Single-Machine Simulation of Federated Learning Systems ([PyTorch](https://github.com/adap/flower/tree/main/examples/simulation-pytorch)) ([Tensorflow](https://github.com/adap/flower/tree/main/examples/simulation-tensorflow))