flwr 1.13.1__py3-none-any.whl → 1.15.0__py3-none-any.whl

flwr/server/driver/grpc_driver.py

@@ -14,38 +14,42 @@
 # ==============================================================================
 """Flower gRPC Driver."""
 
+
 import time
 import warnings
 from collections.abc import Iterable
-from logging import DEBUG, INFO, WARN, WARNING
-from typing import Any, Optional, cast
+from logging import DEBUG, WARNING
+from typing import Optional, cast
 
 import grpc
 
 from flwr.common import DEFAULT_TTL, Message, Metadata, RecordSet
-from flwr.common.constant import MAX_RETRY_DELAY, SERVERAPPIO_API_DEFAULT_CLIENT_ADDRESS
-from flwr.common.grpc import create_channel
+from flwr.common.constant import (
+    SERVERAPPIO_API_DEFAULT_CLIENT_ADDRESS,
+    SUPERLINK_NODE_ID,
+)
+from flwr.common.grpc import create_channel, on_channel_state_change
 from flwr.common.logger import log
-from flwr.common.retry_invoker import RetryInvoker, RetryState, exponential
-from flwr.common.serde import message_from_taskres, message_to_taskins, run_from_proto
+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.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
     GetNodesRequest,
     GetNodesResponse,
-    PullTaskResRequest,
-    PullTaskResResponse,
-    PushTaskInsRequest,
-    PushTaskInsResponse,
+    PullResMessagesRequest,
+    PullResMessagesResponse,
+    PushInsMessagesRequest,
+    PushInsMessagesResponse,
 )
 from flwr.proto.serverappio_pb2_grpc import ServerAppIoStub  # pylint: disable=E0611
-from flwr.proto.task_pb2 import TaskIns  # pylint: disable=E0611
 
 from .driver import Driver
 
 ERROR_MESSAGE_DRIVER_NOT_CONNECTED = """
-[Driver] Error: Not connected.
+[flwr-serverapp] Error: Not connected.
 
 Call `connect()` on the `GrpcDriverStub` instance before calling any of the other
 `GrpcDriverStub` methods.
@@ -75,7 +79,7 @@ class GrpcDriver(Driver):
         self._run: Optional[Run] = None
         self._grpc_stub: Optional[ServerAppIoStub] = None
         self._channel: Optional[grpc.Channel] = None
-        self.node = Node(node_id=0, anonymous=True)
+        self.node = Node(node_id=SUPERLINK_NODE_ID)
         self._retry_invoker = _make_simple_grpc_retry_invoker()
 
     @property
@@ -96,9 +100,10 @@ class GrpcDriver(Driver):
             insecure=(self._cert is None),
             root_certificates=self._cert,
         )
+        self._channel.subscribe(on_channel_state_change)
         self._grpc_stub = ServerAppIoStub(self._channel)
         _wrap_stub(self._grpc_stub, self._retry_invoker)
-        log(DEBUG, "[Driver] Connected to %s", self._addr)
+        log(DEBUG, "[flwr-serverapp] Connected to %s", self._addr)
 
     def _disconnect(self) -> None:
         """Disconnect from the ServerAppIo API."""
@@ -109,7 +114,7 @@ class GrpcDriver(Driver):
         self._channel = None
         self._grpc_stub = None
         channel.close()
-        log(DEBUG, "[Driver] Disconnected")
+        log(DEBUG, "[flwr-serverapp] Disconnected")
 
     def set_run(self, run_id: int) -> None:
         """Set the run."""
@@ -192,20 +197,22 @@ class GrpcDriver(Driver):
         This method takes an iterable of messages and sends each message
         to the node specified in `dst_node_id`.
         """
-        # Construct TaskIns
-        task_ins_list: list[TaskIns] = []
+        # Construct Messages
+        message_proto_list: list[ProtoMessage] = []
         for msg in messages:
             # Check message
             self._check_message(msg)
-            # Convert Message to TaskIns
-            taskins = message_to_taskins(msg)
+            # Convert to proto
+            msg_proto = message_to_proto(msg)
             # Add to list
-            task_ins_list.append(taskins)
+            message_proto_list.append(msg_proto)
         # Call GrpcDriverStub method
-        res: PushTaskInsResponse = self._stub.PushTaskIns(
-            PushTaskInsRequest(task_ins_list=task_ins_list)
+        res: PushInsMessagesResponse = self._stub.PushMessages(
+            PushInsMessagesRequest(
+                messages_list=message_proto_list, run_id=cast(Run, self._run).run_id
+            )
         )
-        return list(res.task_ids)
+        return list(res.message_ids)
 
     def pull_messages(self, message_ids: Iterable[str]) -> Iterable[Message]:
         """Pull messages based on message IDs.
@@ -213,12 +220,15 @@ class GrpcDriver(Driver):
         This method is used to collect messages from the SuperLink that correspond to a
         set of given message IDs.
         """
-        # Pull TaskRes
-        res: PullTaskResResponse = self._stub.PullTaskRes(
-            PullTaskResRequest(node=self.node, task_ids=message_ids)
+        # Pull Messages
+        res: PullResMessagesResponse = self._stub.PullMessages(
+            PullResMessagesRequest(
+                message_ids=message_ids,
+                run_id=cast(Run, self._run).run_id,
+            )
         )
-        # Convert TaskRes to Message
-        msgs = [message_from_taskres(taskres) for taskres in res.task_res_list]
+        # Convert Message from Protobuf representation
+        msgs = [message_from_proto(msg_proto) for msg_proto in res.messages_list]
         return msgs
 
     def send_and_receive(
@@ -258,60 +268,3 @@ class GrpcDriver(Driver):
             return
         # Disconnect
         self._disconnect()
-
-
-def _make_simple_grpc_retry_invoker() -> RetryInvoker:
-    """Create a simple gRPC retry invoker."""
-
-    def _on_sucess(retry_state: RetryState) -> None:
-        if retry_state.tries > 1:
-            log(
-                INFO,
-                "Connection successful after %.2f seconds and %s tries.",
-                retry_state.elapsed_time,
-                retry_state.tries,
-            )
-
-    def _on_backoff(retry_state: RetryState) -> None:
-        if retry_state.tries == 1:
-            log(WARN, "Connection attempt failed, retrying...")
-        else:
-            log(
-                WARN,
-                "Connection attempt failed, retrying in %.2f seconds",
-                retry_state.actual_wait,
-            )
-
-    def _on_giveup(retry_state: RetryState) -> None:
-        if retry_state.tries > 1:
-            log(
-                WARN,
-                "Giving up reconnection after %.2f seconds and %s tries.",
-                retry_state.elapsed_time,
-                retry_state.tries,
-            )
-
-    return RetryInvoker(
-        wait_gen_factory=lambda: exponential(max_delay=MAX_RETRY_DELAY),
-        recoverable_exceptions=grpc.RpcError,
-        max_tries=None,
-        max_time=None,
-        on_success=_on_sucess,
-        on_backoff=_on_backoff,
-        on_giveup=_on_giveup,
-        should_giveup=lambda e: e.code() != grpc.StatusCode.UNAVAILABLE,  # type: ignore
-    )
-
-
-def _wrap_stub(stub: ServerAppIoStub, retry_invoker: RetryInvoker) -> None:
-    """Wrap the gRPC stub with a retry invoker."""
-
-    def make_lambda(original_method: Any) -> Any:
-        return lambda *args, **kwargs: retry_invoker.invoke(
-            original_method, *args, **kwargs
-        )
-
-    for method_name in vars(stub):
-        method = getattr(stub, method_name)
-        if callable(method):
-            setattr(stub, method_name, make_lambda(method))

flwr/server/driver/inmemory_driver.py

@@ -22,6 +22,7 @@ from typing import Optional, cast
 from uuid import UUID
 
 from flwr.common import DEFAULT_TTL, Message, Metadata, RecordSet
+from flwr.common.constant import SUPERLINK_NODE_ID
 from flwr.common.serde import message_from_taskres, message_to_taskins
 from flwr.common.typing import Run
 from flwr.proto.node_pb2 import Node  # pylint: disable=E0611
@@ -49,7 +50,7 @@ class InMemoryDriver(Driver):
         self._run: Optional[Run] = None
         self.state = state_factory.state()
         self.pull_interval = pull_interval
-        self.node = Node(node_id=0, anonymous=True)
+        self.node = Node(node_id=SUPERLINK_NODE_ID)
 
     def _check_message(self, message: Message) -> None:
         # Check if the message is valid
@@ -142,7 +143,11 @@ class InMemoryDriver(Driver):
         # Pull TaskRes
         task_res_list = self.state.get_task_res(task_ids=msg_ids)
         # Delete tasks in state
-        self.state.delete_tasks(msg_ids)
+        # Delete the TaskIns/TaskRes pairs if TaskRes is found
+        task_ins_ids_to_delete = {
+            UUID(task_res.task.ancestry[0]) for task_res in task_res_list
+        }
+        self.state.delete_tasks(task_ins_ids=task_ins_ids_to_delete)
         # Convert TaskRes to Message
         msgs = [message_from_taskres(taskres) for taskres in task_res_list]
         return msgs

flwr/server/run_serverapp.py

@@ -15,12 +15,12 @@
 """Run ServerApp."""
 
 
-import sys
 from logging import DEBUG, ERROR
 from typing import Optional
 
-from flwr.common import Context
-from flwr.common.logger import log, warn_unsupported_feature
+from flwr.common import Context, EventType, event
+from flwr.common.exit_handlers import register_exit_handlers
+from flwr.common.logger import log
 from flwr.common.object_ref import load_app
 
 from .driver import Driver
@@ -66,12 +66,11 @@ def run(
     return context
 
 
-# pylint: disable-next=too-many-branches,too-many-statements,too-many-locals
 def run_server_app() -> None:
     """Run Flower server app."""
-    warn_unsupported_feature(
-        "The command `flower-server-app` is deprecated and no longer in use. "
-        "Use the `flwr-serverapp` exclusively instead."
+    event(EventType.RUN_SERVER_APP_ENTER)
+    log(
+        ERROR,
+        "The command `flower-server-app` has been replaced by `flwr run`.",
     )
-    log(ERROR, "`flower-server-app` used.")
-    sys.exit()
+    register_exit_handlers(event_type=EventType.RUN_SERVER_APP_LEAVE)

flwr/server/serverapp/app.py

@@ -14,8 +14,8 @@
 # ==============================================================================
 """Flower ServerApp process."""
 
+
 import argparse
-import sys
 from logging import DEBUG, ERROR, INFO
 from pathlib import Path
 from queue import Queue
@@ -24,6 +24,7 @@ from typing import Optional
 
 from flwr.cli.config_utils import get_fab_metadata
 from flwr.cli.install import install_from_fab
+from flwr.cli.utils import get_sha256_hash
 from flwr.common.args import add_args_flwr_app_common
 from flwr.common.config import (
     get_flwr_dir,
@@ -36,6 +37,7 @@ from flwr.common.constant import (
     Status,
     SubStatus,
 )
+from flwr.common.exit import ExitCode, flwr_exit
 from flwr.common.logger import (
     log,
     mirror_output_to_queue,
@@ -50,7 +52,8 @@ from flwr.common.serde import (
     run_from_proto,
     run_status_to_proto,
 )
-from flwr.common.typing import RunStatus
+from flwr.common.telemetry import EventType, event
+from flwr.common.typing import RunNotRunningException, RunStatus
 from flwr.proto.run_pb2 import UpdateRunStatusRequest  # pylint: disable=E0611
 from flwr.proto.serverappio_pb2 import (  # pylint: disable=E0611
     PullServerAppInputsRequest,
@@ -69,19 +72,18 @@ def flwr_serverapp() -> None:
 
     args = _parse_args_run_flwr_serverapp().parse_args()
 
-    log(INFO, "Starting Flower ServerApp")
+    log(INFO, "Start `flwr-serverapp` process")
 
     if not args.insecure:
-        log(
-            ERROR,
-            "`flwr-serverapp` does not support TLS yet. "
-            "Please use the '--insecure' flag.",
+        flwr_exit(
+            ExitCode.COMMON_TLS_NOT_SUPPORTED,
+            "`flwr-serverapp` does not support TLS yet.",
         )
-        sys.exit(1)
 
     log(
         DEBUG,
-        "Starting isolated `ServerApp` connected to SuperLink's ServerAppIo API at %s",
+        "`flwr-serverapp` will attempt to connect to SuperLink's "
+        "ServerAppIo API at %s",
         args.serverappio_api_address,
     )
     run_serverapp(
@@ -96,7 +98,7 @@ def flwr_serverapp() -> None:
     restore_output()
 
 
-def run_serverapp(  # pylint: disable=R0914, disable=W0212
+def run_serverapp(  # pylint: disable=R0914, disable=W0212, disable=R0915
     serverappio_api_address: str,
     log_queue: Queue[Optional[str]],
     run_once: bool,
@@ -112,12 +114,15 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
     # Resolve directory where FABs are installed
     flwr_dir_ = get_flwr_dir(flwr_dir)
     log_uploader = None
-
+    success = True
+    hash_run_id = None
+    run_status = None
     while True:
 
         try:
             # Pull ServerAppInputs from LinkState
             req = PullServerAppInputsRequest()
+            log(DEBUG, "[flwr-serverapp] Pull ServerAppInputs")
             res: PullServerAppInputsResponse = driver._stub.PullServerAppInputs(req)
             if not res.HasField("run"):
                 sleep(3)
@@ -128,6 +133,8 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
             run = run_from_proto(res.run)
             fab = fab_from_proto(res.fab)
 
+            hash_run_id = get_sha256_hash(run.run_id)
+
             driver.set_run(run.run_id)
 
             # Start log uploader for this run
@@ -138,7 +145,7 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
                 stub=driver._stub,
             )
 
-            log(DEBUG, "ServerApp process starts FAB installation.")
+            log(DEBUG, "[flwr-serverapp] Start FAB installation.")
             install_from_fab(fab.content, flwr_dir=flwr_dir_, skip_prompt=True)
 
             fab_id, fab_version = get_fab_metadata(fab.content)
@@ -159,7 +166,7 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
 
             log(
                 DEBUG,
-                "Flower will load ServerApp `%s` in %s",
+                "[flwr-serverapp] Will load ServerApp `%s` in %s",
                 server_app_attr,
                 app_path,
             )
@@ -170,6 +177,11 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
                 UpdateRunStatusRequest(run_id=run.run_id, run_status=run_status_proto)
             )
 
+            event(
+                EventType.FLWR_SERVERAPP_RUN_ENTER,
+                event_details={"run-id-hash": hash_run_id},
+            )
+
             # Load and run the ServerApp with the Driver
             updated_context = run_(
                 driver=driver,
@@ -180,17 +192,25 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
 
             # Send resulting context
             context_proto = context_to_proto(updated_context)
+            log(DEBUG, "[flwr-serverapp] Will push ServerAppOutputs")
             out_req = PushServerAppOutputsRequest(
                 run_id=run.run_id, context=context_proto
             )
             _ = driver._stub.PushServerAppOutputs(out_req)
 
             run_status = RunStatus(Status.FINISHED, SubStatus.COMPLETED, "")
+        except RunNotRunningException:
+            log(INFO, "")
+            log(INFO, "Run ID %s stopped.", run.run_id)
+            log(INFO, "")
+            run_status = None
+            success = False
 
         except Exception as ex:  # pylint: disable=broad-exception-caught
             exc_entity = "ServerApp"
             log(ERROR, "%s raised an exception", exc_entity, exc_info=ex)
             run_status = RunStatus(Status.FINISHED, SubStatus.FAILED, str(ex))
+            success = False
 
         finally:
             # Stop log uploader for this run and upload final logs
@@ -206,6 +226,10 @@ def run_serverapp(  # pylint: disable=R0914, disable=W0212
                         run_id=run.run_id, run_status=run_status_proto
                     )
                 )
+            event(
+                EventType.FLWR_SERVERAPP_RUN_LEAVE,
+                event_details={"run-id-hash": hash_run_id, "success": success},
+            )
 
         # Stop the loop if `flwr-serverapp` is expected to process a single run
         if run_once:

flwr/server/strategy/dpfedavg_fixed.py

@@ -17,6 +17,7 @@
 Paper: arxiv.org/pdf/1710.06963.pdf
 """
 
+
 from typing import Optional, Union
 
 from flwr.common import EvaluateIns, EvaluateRes, FitIns, FitRes, Parameters, Scalar

flwr/server/superlink/driver/serverappio_grpc.py

@@ -14,12 +14,14 @@
 # ==============================================================================
 """ServerAppIo gRPC API."""
 
+
 from logging import INFO
 from typing import Optional
 
 import grpc
 
 from flwr.common import GRPC_MAX_MESSAGE_LENGTH
+from flwr.common.grpc import generic_create_grpc_server
 from flwr.common.logger import log
 from flwr.proto.serverappio_pb2_grpc import (  # pylint: disable=E0611
     add_ServerAppIoServicer_to_server,
@@ -27,7 +29,6 @@ from flwr.proto.serverappio_pb2_grpc import (  # pylint: disable=E0611
 from flwr.server.superlink.ffs.ffs_factory import FfsFactory
 from flwr.server.superlink.linkstate import LinkStateFactory
 
-from ..fleet.grpc_bidi.grpc_server import generic_create_grpc_server
 from .serverappio_servicer import ServerAppIoServicer