Merge branch 'main' into update-hf-template

src/py/flwr/simulation/run_simulation.py

@@ -216,6 +216,7 @@ def run_simulation_from_cli() -> None:
         app_dir=app_dir,
         run=run,
         enable_tf_gpu_growth=args.enable_tf_gpu_growth,
+        delay_start=args.delay_start,
         verbose_logging=args.verbose,
         server_app_run_config=fused_config,
         is_app=is_app,
@@ -309,7 +310,6 @@ def run_serverapp_th(
     f_stop: threading.Event,
     has_exception: threading.Event,
     enable_tf_gpu_growth: bool,
-    delay_launch: int = 3,
 ) -> threading.Thread:
     """Run SeverApp in a thread."""
 
@@ -365,7 +365,6 @@ def server_th_with_start_checks(
             server_app,
         ),
     )
-    sleep(delay_launch)
     serverapp_th.start()
     return serverapp_th
 
@@ -380,6 +379,7 @@ def _main_loop(
     enable_tf_gpu_growth: bool,
     run: Run,
     exit_event: EventType,
+    delay_start: int,
     flwr_dir: Optional[str] = None,
     client_app: Optional[ClientApp] = None,
     client_app_attr: Optional[str] = None,
@@ -419,6 +419,9 @@ def _main_loop(
             enable_tf_gpu_growth=enable_tf_gpu_growth,
         )
 
+        # Buffer time so the `ServerApp` in separate thread is ready
+        log(DEBUG, "Buffer time delay: %ds", delay_start)
+        sleep(delay_start)
         # Start Simulation Engine
         vce.start_vce(
             num_supernodes=num_supernodes,
@@ -467,6 +470,7 @@ def _run_simulation(
     flwr_dir: Optional[str] = None,
     run: Optional[Run] = None,
     enable_tf_gpu_growth: bool = False,
+    delay_start: int = 5,
     verbose_logging: bool = False,
     is_app: bool = False,
 ) -> None:
@@ -523,6 +527,7 @@ def _run_simulation(
         enable_tf_gpu_growth,
         run,
         exit_event,
+        delay_start,
         flwr_dir,
         client_app,
         client_app_attr,
@@ -610,6 +615,13 @@ def _parse_args_run_simulation() -> argparse.ArgumentParser:
         "Read more about how `tf.config.experimental.set_memory_growth()` works in "
         "the TensorFlow documentation: https://www.tensorflow.org/api/stable.",
     )
+    parser.add_argument(
+        "--delay-start",
+        type=int,
+        default=3,
+        help="Buffer time (in seconds) to delay the start the simulation engine after "
+        "the `ServerApp`, which runs in a separate thread, has been launched.",
+    )
     parser.add_argument(
         "--verbose",
         action="store_true",