flwr-nightly 1.19.0.dev20250603__tar.gz → 1.19.0.dev20250605__tar.gz

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: flwr-nightly
-Version: 1.19.0.dev20250603
+Version: 1.19.0.dev20250605
 Summary: Flower: A Friendly Federated AI Framework
 License: Apache-2.0
 Keywords: Artificial Intelligence,Federated AI,Federated Analytics,Federated Evaluation,Federated Learning,Flower,Machine Learning

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/client/grpc_rere_client/connection.py

@@ -14,11 +14,10 @@
 # ==============================================================================
 """Contextmanager for a gRPC request-response channel to the Flower server."""
 
-
 from collections.abc import Iterator, Sequence
 from contextlib import contextmanager
 from copy import copy
-from logging import ERROR
+from logging import DEBUG, ERROR
 from pathlib import Path
 from typing import Callable, Optional, Union, cast
 
@@ -31,13 +30,21 @@ from flwr.common import GRPC_MAX_MESSAGE_LENGTH
 from flwr.common.constant import HEARTBEAT_CALL_TIMEOUT, HEARTBEAT_DEFAULT_INTERVAL
 from flwr.common.grpc import create_channel, on_channel_state_change
 from flwr.common.heartbeat import HeartbeatSender
+from flwr.common.inflatable_grpc_utils import (
+    pull_object_from_servicer,
+    push_object_to_servicer,
+)
 from flwr.common.logger import log
-from flwr.common.message import Message
+from flwr.common.message import (
+    Message,
+    get_message_to_descendant_id_mapping,
+    remove_content_from_message,
+)
 from flwr.common.retry_invoker import RetryInvoker, _wrap_stub
 from flwr.common.secure_aggregation.crypto.symmetric_encryption import (
     generate_key_pairs,
 )
-from flwr.common.serde import message_from_proto, message_to_proto, run_from_proto
+from flwr.common.serde import message_to_proto, run_from_proto
 from flwr.common.typing import Fab, Run, RunNotRunningException
 from flwr.proto.fab_pb2 import GetFabRequest, GetFabResponse  # pylint: disable=E0611
 from flwr.proto.fleet_pb2 import (  # pylint: disable=E0611
@@ -46,6 +53,7 @@ from flwr.proto.fleet_pb2 import (  # pylint: disable=E0611
     PullMessagesRequest,
     PullMessagesResponse,
     PushMessagesRequest,
+    PushMessagesResponse,
 )
 from flwr.proto.fleet_pb2_grpc import FleetStub  # pylint: disable=E0611
 from flwr.proto.heartbeat_pb2 import (  # pylint: disable=E0611
@@ -254,7 +262,24 @@ def grpc_request_response(  # pylint: disable=R0913,R0914,R0915,R0917
             message_proto = None
 
         # Construct the Message
-        in_message = message_from_proto(message_proto) if message_proto else None
+        in_message: Optional[Message] = None
+
+        if message_proto:
+            in_message = cast(
+                Message,
+                pull_object_from_servicer(
+                    object_id=message_proto.metadata.message_id,
+                    stub=stub,
+                    node=node,
+                    run_id=message_proto.metadata.run_id,
+                ),
+            )
+
+        if in_message:
+            # The deflated message doesn't contain the message_id (its own object_id)
+            # Inject
+            # pylint: disable-next=W0212
+            in_message.metadata._message_id = message_proto.metadata.message_id  # type: ignore
 
         # Remember `metadata` of the in message
         nonlocal metadata
@@ -284,9 +309,25 @@ def grpc_request_response(  # pylint: disable=R0913,R0914,R0915,R0917
             return
 
         # Serialize Message
-        message_proto = message_to_proto(message=message)
-        request = PushMessagesRequest(node=node, messages_list=[message_proto])
-        _ = stub.PushMessages(request)
+        message_proto = message_to_proto(message=remove_content_from_message(message))
+        descendants_mapping = get_message_to_descendant_id_mapping(message)
+        request = PushMessagesRequest(
+            node=node,
+            messages_list=[message_proto],
+            msg_to_descendant_mapping=descendants_mapping,
+        )
+        response: PushMessagesResponse = stub.PushMessages(request=request)
+
+        if response.objects_to_push:
+            objs_to_push = set(response.objects_to_push[message.object_id].object_ids)
+            push_object_to_servicer(
+                message,
+                stub,
+                node,
+                run_id=message.metadata.run_id,
+                object_ids_to_push=objs_to_push,
+            )
+            log(DEBUG, "Pushed %s objects to servicer.", len(objs_to_push))
 
         # Cleanup
         metadata = None

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/common/inflatable_grpc_utils.py

@@ -15,6 +15,7 @@
 """InflatableObject utils."""
 
 
+from time import sleep
 from typing import Optional, Union
 
 from flwr.client.grpc_rere_client.grpc_adapter import GrpcAdapter
@@ -94,10 +95,14 @@ def pull_object_from_servicer(
 ) -> InflatableObject:
     """Recursively inflate an object by pulling it from the servicer."""
     # Pull object
-    object_proto: PullObjectResponse = stub.PullObject(
-        PullObjectRequest(node=node, run_id=run_id, object_id=object_id)
-    )
-    object_content = object_proto.object_content
+    object_available = False
+    while not object_available:
+        object_proto: PullObjectResponse = stub.PullObject(
+            PullObjectRequest(node=node, run_id=run_id, object_id=object_id)
+        )
+        object_available = object_proto.object_available
+        object_content = object_proto.object_content
+        sleep(0.1)
 
     # Extract object class and object_ids of children
     obj_type, children_obj_ids, _ = get_object_head_values_from_object_content(

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/common/message.py

@@ -426,6 +426,17 @@ def make_message(
     return Message(metadata=metadata, content=content, error=error)  # type: ignore
 
 
+def remove_content_from_message(message: Message) -> Message:
+    """Return a copy of the Message but with an empty RecordDict as content.
+
+    If message has no content, it returns itself.
+    """
+    if message.has_error():
+        return message
+
+    return make_message(metadata=message.metadata, content=RecordDict())
+
+
 def _limit_reply_ttl(
     current: float, reply_ttl: float | None, reply_to: Message
 ) -> float:

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/server/compat/app_utils.py

@@ -16,6 +16,7 @@
 
 
 import threading
+from typing import Any, Callable
 
 from flwr.common.typing import RunNotRunningException
 
@@ -80,36 +81,57 @@ def _update_client_manager(
     """Update the nodes list in the client manager."""
     # Loop until the grid is disconnected
     registered_nodes: dict[int, GridClientProxy] = {}
+    lock = threading.RLock()
+
+    def update_registered_nodes() -> None:
+        with lock:
+            all_node_ids = set(grid.get_node_ids())
+            dead_nodes = set(registered_nodes).difference(all_node_ids)
+            new_nodes = all_node_ids.difference(registered_nodes)
+
+            # Unregister dead nodes
+            for node_id in dead_nodes:
+                client_proxy = registered_nodes[node_id]
+                client_manager.unregister(client_proxy)
+                del registered_nodes[node_id]
+
+            # Register new nodes
+            for node_id in new_nodes:
+                client_proxy = GridClientProxy(
+                    node_id=node_id,
+                    grid=grid,
+                    run_id=grid.run.run_id,
+                )
+                if client_manager.register(client_proxy):
+                    registered_nodes[node_id] = client_proxy
+                else:
+                    raise RuntimeError("Could not register node.")
+
+    # Get the wrapped method of ClientManager instance
+    def get_wrapped_method(method_name: str) -> Callable[..., Any]:
+        original_method = getattr(client_manager, method_name)
+
+        def wrapped_method(*args: Any, **kwargs: Any) -> Any:
+            # Update registered nodes before calling the original method
+            update_registered_nodes()
+            return original_method(*args, **kwargs)
+
+        return wrapped_method
+
+    # Wrap the ClientManager
+    for method_name in ["num_available", "all", "sample"]:
+        setattr(client_manager, method_name, get_wrapped_method(method_name))
+
+    c_done.set()
+
     while not f_stop.is_set():
+        # Sleep for 5 seconds
+        if not f_stop.is_set():
+            f_stop.wait(5)
+
         try:
-            all_node_ids = set(grid.get_node_ids())
+            # Update registered nodes
+            update_registered_nodes()
         except RunNotRunningException:
             f_stop.set()
             break
-        dead_nodes = set(registered_nodes).difference(all_node_ids)
-        new_nodes = all_node_ids.difference(registered_nodes)
-
-        # Unregister dead nodes
-        for node_id in dead_nodes:
-            client_proxy = registered_nodes[node_id]
-            client_manager.unregister(client_proxy)
-            del registered_nodes[node_id]
-
-        # Register new nodes
-        for node_id in new_nodes:
-            client_proxy = GridClientProxy(
-                node_id=node_id,
-                grid=grid,
-                run_id=grid.run.run_id,
-            )
-            if client_manager.register(client_proxy):
-                registered_nodes[node_id] = client_proxy
-            else:
-                raise RuntimeError("Could not register node.")
-
-        # Flag first pass for nodes registration is completed
-        c_done.set()
-
-        # Sleep for 3 seconds
-        if not f_stop.is_set():
-            f_stop.wait(3)

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/server/grid/grpc_grid.py

@@ -28,11 +28,18 @@ from flwr.common.constant import (
     SUPERLINK_NODE_ID,
 )
 from flwr.common.grpc import create_channel, on_channel_state_change
+from flwr.common.inflatable_grpc_utils import (
+    pull_object_from_servicer,
+    push_object_to_servicer,
+)
 from flwr.common.logger import log, warn_deprecated_feature
+from flwr.common.message import (
+    get_message_to_descendant_id_mapping,
+    remove_content_from_message,
+)
 from flwr.common.retry_invoker import _make_simple_grpc_retry_invoker, _wrap_stub
-from flwr.common.serde import message_from_proto, message_to_proto, run_from_proto
+from flwr.common.serde import message_to_proto, run_from_proto
 from flwr.common.typing import Run
-from flwr.proto.message_pb2 import Message as ProtoMessage  # pylint: disable=E0611
 from flwr.proto.node_pb2 import Node  # pylint: disable=E0611
 from flwr.proto.run_pb2 import GetRunRequest, GetRunResponse  # pylint: disable=E0611
 from flwr.proto.serverappio_pb2 import (  # pylint: disable=E0611
@@ -198,6 +205,35 @@ class GrpcGrid(Grid):
         )
         return [node.node_id for node in res.nodes]
 
+    def _try_push_message(self, run_id: int, message: Message) -> str:
+        """Push one message and its associated objects."""
+        # Compute mapping of message descendants
+        descendants_mapping = get_message_to_descendant_id_mapping(message)
+
+        # Call GrpcServerAppIoStub method
+        res: PushInsMessagesResponse = self._stub.PushMessages(
+            PushInsMessagesRequest(
+                messages_list=[message_to_proto(remove_content_from_message(message))],
+                run_id=run_id,
+                msg_to_descendant_mapping=descendants_mapping,
+            )
+        )
+
+        # Push objects
+        msg_id = res.message_ids[0]
+        # If Message was added to the LinkState correctly
+        if msg_id is not None:
+            obj_ids_to_push = set(res.objects_to_push[msg_id].object_ids)
+            # Push only object that are not in the store
+            push_object_to_servicer(
+                message,
+                self._stub,
+                node=self.node,
+                run_id=run_id,
+                object_ids_to_push=obj_ids_to_push,
+            )
+        return msg_id
+
     def push_messages(self, messages: Iterable[Message]) -> Iterable[str]:
         """Push messages to specified node IDs.
 
@@ -206,58 +242,64 @@ class GrpcGrid(Grid):
         """
         # Construct Messages
         run_id = cast(Run, self._run).run_id
-        message_proto_list: list[ProtoMessage] = []
-        for msg in messages:
-            # Populate metadata
-            msg.metadata.__dict__["_run_id"] = run_id
-            msg.metadata.__dict__["_src_node_id"] = self.node.node_id
-            msg.metadata.__dict__["_message_id"] = msg.object_id
-            # Check message
-            self._check_message(msg)
-            # Convert to proto
-            msg_proto = message_to_proto(msg)
-            # Add to list
-            message_proto_list.append(msg_proto)
-
+        message_ids: list[str] = []
         try:
-            # Call GrpcServerAppIoStub method
-            res: PushInsMessagesResponse = self._stub.PushMessages(
-                PushInsMessagesRequest(messages_list=message_proto_list, run_id=run_id)
-            )
-            if len([msg_id for msg_id in res.message_ids if msg_id]) != len(
-                message_proto_list
-            ):
-                log(
-                    WARNING,
-                    "Not all messages could be pushed to the SuperLink. The returned "
-                    "list has `None` for those messages (the order is preserved as "
-                    "passed to `push_messages`). This could be due to a malformed "
-                    "message.",
-                )
-            return list(res.message_ids)
+            for msg in messages:
+                # Populate metadata
+                msg.metadata.__dict__["_run_id"] = run_id
+                msg.metadata.__dict__["_src_node_id"] = self.node.node_id
+                msg.metadata.__dict__["_message_id"] = msg.object_id
+                # Check message
+                self._check_message(msg)
+                # Try pushing message and its objects
+                message_ids.append(self._try_push_message(run_id, msg))
+
         except grpc.RpcError as e:
             if e.code() == grpc.StatusCode.RESOURCE_EXHAUSTED:  # pylint: disable=E1101
                 log(ERROR, ERROR_MESSAGE_PUSH_MESSAGES_RESOURCE_EXHAUSTED)
                 return []
             raise
 
+        if None in message_ids:
+            log(
+                WARNING,
+                "Not all messages could be pushed to the SuperLink. The returned "
+                "list has `None` for those messages (the order is preserved as "
+                "passed to `push_messages`). This could be due to a malformed "
+                "message.",
+            )
+
+        return message_ids
+
     def pull_messages(self, message_ids: Iterable[str]) -> Iterable[Message]:
         """Pull messages based on message IDs.
 
         This method is used to collect messages from the SuperLink that correspond to a
         set of given message IDs.
         """
+        run_id = cast(Run, self._run).run_id
         try:
             # Pull Messages
             res: PullResMessagesResponse = self._stub.PullMessages(
                 PullResMessagesRequest(
                     message_ids=message_ids,
-                    run_id=cast(Run, self._run).run_id,
+                    run_id=run_id,
                 )
             )
-            # Convert Message from Protobuf representation
-            msgs = [message_from_proto(msg_proto) for msg_proto in res.messages_list]
-            return msgs
+            # Pull Messages from store
+            inflated_msgs: list[Message] = []
+            for msg_proto in res.messages_list:
+
+                message = pull_object_from_servicer(
+                    msg_proto.metadata.message_id,
+                    self._stub,
+                    node=self.node,
+                    run_id=run_id,
+                )
+                inflated_msgs.append(cast(Message, message))
+
+            return inflated_msgs
+
         except grpc.RpcError as e:
             if e.code() == grpc.StatusCode.RESOURCE_EXHAUSTED:  # pylint: disable=E1101
                 log(ERROR, ERROR_MESSAGE_PULL_MESSAGES_RESOURCE_EXHAUSTED)

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/server/superlink/ffs/__init__.py

@@ -17,8 +17,10 @@
 
 from .disk_ffs import DiskFfs as DiskFfs
 from .ffs import Ffs as Ffs
+from .ffs_factory import FfsFactory as FfsFactory
 
 __all__ = [
     "DiskFfs",
     "Ffs",
+    "FfsFactory",
 ]

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/supernode/cli/flwr_clientapp.py

@@ -48,6 +48,7 @@ def flwr_clientapp() -> None:
         token=args.token,
         flwr_dir=args.flwr_dir,
         certificates=None,
+        parent_pid=args.parent_pid,
     )
 
 
@@ -69,5 +70,12 @@ def _parse_args_run_flwr_clientapp() -> argparse.ArgumentParser:
         required=False,
         help="Unique token generated by SuperNode for each ClientApp execution",
     )
+    parser.add_argument(
+        "--parent-pid",
+        type=int,
+        default=None,
+        help="The PID of the parent process. When set, the process will terminate "
+        "when the parent process exits.",
+    )
     add_args_flwr_app_common(parser=parser)
     return parser

{flwr_nightly-1.19.0.dev20250603 → flwr_nightly-1.19.0.dev20250605}/py/flwr/supernode/runtime/run_clientapp.py

@@ -16,6 +16,8 @@
 
 
 import gc
+import os
+import threading
 import time
 from logging import DEBUG, ERROR, INFO
 from typing import Optional
@@ -54,14 +56,19 @@ from flwr.proto.clientappio_pb2 import (
 from flwr.proto.clientappio_pb2_grpc import ClientAppIoStub
 
 
-def run_clientapp(  # pylint: disable=R0914
+def run_clientapp(  # pylint: disable=R0913, R0914, R0917
     clientappio_api_address: str,
     run_once: bool,
     token: Optional[int] = None,
     flwr_dir: Optional[str] = None,
     certificates: Optional[bytes] = None,
+    parent_pid: Optional[int] = None,
 ) -> None:
     """Run Flower ClientApp process."""
+    # Monitor the main process in case of SIGKILL
+    if parent_pid is not None:
+        start_parent_process_monitor(parent_pid)
+
     channel = create_channel(
         server_address=clientappio_api_address,
         insecure=(certificates is None),
@@ -151,6 +158,20 @@ def run_clientapp(  # pylint: disable=R0914
         channel.close()
 
 
+def start_parent_process_monitor(
+    parent_pid: int,
+) -> None:
+    """Monitor the parent process and exit if it terminates."""
+
+    def monitor() -> None:
+        while True:
+            time.sleep(0.2)
+            if os.getppid() != parent_pid:
+                os.kill(os.getpid(), 9)
+
+    threading.Thread(target=monitor, daemon=True).start()
+
+
 def get_token(stub: grpc.Channel) -> Optional[int]:
     """Get a token from SuperNode."""
     log(DEBUG, "[flwr-clientapp] Request token")