flwr-nightly 1.8.0.dev20240315__py3-none-any.whl → 1.8.0.dev20240327__py3-none-any.whl

flwr/server/superlink/fleet/message_handler/message_handler.py

@@ -15,6 +15,7 @@
 """Fleet API message handlers."""
 
 
+import time
 from typing import List, Optional
 from uuid import UUID
 
@@ -23,6 +24,8 @@ from flwr.proto.fleet_pb2 import (  # pylint: disable=E0611
     CreateNodeResponse,
     DeleteNodeRequest,
     DeleteNodeResponse,
+    PingRequest,
+    PingResponse,
     PullTaskInsRequest,
     PullTaskInsResponse,
     PushTaskResRequest,
@@ -55,6 +58,14 @@ def delete_node(request: DeleteNodeRequest, state: State) -> DeleteNodeResponse:
     return DeleteNodeResponse()
 
 
+def ping(
+    request: PingRequest,  # pylint: disable=unused-argument
+    state: State,  # pylint: disable=unused-argument
+) -> PingResponse:
+    """."""
+    return PingResponse(success=True)
+
+
 def pull_task_ins(request: PullTaskInsRequest, state: State) -> PullTaskInsResponse:
     """Pull TaskIns handler."""
     # Get node_id if client node is not anonymous
@@ -77,6 +88,9 @@ def push_task_res(request: PushTaskResRequest, state: State) -> PushTaskResRespo
     task_res: TaskRes = request.task_res_list[0]
     # pylint: enable=no-member
 
+    # Set pushed_at (timestamp in seconds)
+    task_res.task.pushed_at = time.time()
+
     # Store TaskRes in State
     task_id: Optional[UUID] = state.store_task_res(task_res=task_res)
 

flwr/server/superlink/state/in_memory_state.py

@@ -17,9 +17,9 @@
 
 import os
 import threading
-from datetime import datetime, timedelta
+import time
 from logging import ERROR
-from typing import Dict, List, Optional, Set
+from typing import Dict, List, Optional, Set, Tuple
 from uuid import UUID, uuid4
 
 from flwr.common import log, now
@@ -32,7 +32,8 @@ class InMemoryState(State):
     """In-memory State implementation."""
 
     def __init__(self) -> None:
-        self.node_ids: Set[int] = set()
+        # Map node_id to (online_until, ping_interval)
+        self.node_ids: Dict[int, Tuple[float, float]] = {}
         self.run_ids: Set[int] = set()
         self.task_ins_store: Dict[UUID, TaskIns] = {}
         self.task_res_store: Dict[UUID, TaskRes] = {}
@@ -50,15 +51,11 @@ class InMemoryState(State):
             log(ERROR, "`run_id` is invalid")
             return None
 
-        # Create task_id, created_at and ttl
+        # Create task_id
         task_id = uuid4()
-        created_at: datetime = now()
-        ttl: datetime = created_at + timedelta(hours=24)
 
         # Store TaskIns
         task_ins.task_id = str(task_id)
-        task_ins.task.created_at = created_at.isoformat()
-        task_ins.task.ttl = ttl.isoformat()
         with self.lock:
             self.task_ins_store[task_id] = task_ins
 
@@ -113,15 +110,11 @@ class InMemoryState(State):
             log(ERROR, "`run_id` is invalid")
             return None
 
-        # Create task_id, created_at and ttl
+        # Create task_id
         task_id = uuid4()
-        created_at: datetime = now()
-        ttl: datetime = created_at + timedelta(hours=24)
 
         # Store TaskRes
         task_res.task_id = str(task_id)
-        task_res.task.created_at = created_at.isoformat()
-        task_res.task.ttl = ttl.isoformat()
         with self.lock:
             self.task_res_store[task_id] = task_res
 
@@ -194,17 +187,21 @@ class InMemoryState(State):
         # Sample a random int64 as node_id
         node_id: int = int.from_bytes(os.urandom(8), "little", signed=True)
 
-        if node_id not in self.node_ids:
-            self.node_ids.add(node_id)
-            return node_id
+        with self.lock:
+            if node_id not in self.node_ids:
+                # Default ping interval is 30s
+                # TODO: change 1e9 to 30s  # pylint: disable=W0511
+                self.node_ids[node_id] = (time.time() + 1e9, 1e9)
+                return node_id
         log(ERROR, "Unexpected node registration failure.")
         return 0
 
     def delete_node(self, node_id: int) -> None:
         """Delete a client node."""
-        if node_id not in self.node_ids:
-            raise ValueError(f"Node {node_id} not found")
-        self.node_ids.remove(node_id)
+        with self.lock:
+            if node_id not in self.node_ids:
+                raise ValueError(f"Node {node_id} not found")
+            del self.node_ids[node_id]
 
     def get_nodes(self, run_id: int) -> Set[int]:
         """Return all available client nodes.
@@ -214,17 +211,32 @@ class InMemoryState(State):
         If the provided `run_id` does not exist or has no matching nodes,
         an empty `Set` MUST be returned.
         """
-        if run_id not in self.run_ids:
-            return set()
-        return self.node_ids
+        with self.lock:
+            if run_id not in self.run_ids:
+                return set()
+            current_time = time.time()
+            return {
+                node_id
+                for node_id, (online_until, _) in self.node_ids.items()
+                if online_until > current_time
+            }
 
     def create_run(self) -> int:
         """Create one run."""
         # Sample a random int64 as run_id
-        run_id: int = int.from_bytes(os.urandom(8), "little", signed=True)
+        with self.lock:
+            run_id: int = int.from_bytes(os.urandom(8), "little", signed=True)
 
-        if run_id not in self.run_ids:
-            self.run_ids.add(run_id)
-            return run_id
+            if run_id not in self.run_ids:
+                self.run_ids.add(run_id)
+                return run_id
         log(ERROR, "Unexpected run creation failure.")
         return 0
+
+    def acknowledge_ping(self, node_id: int, ping_interval: float) -> bool:
+        """Acknowledge a ping received from a node, serving as a heartbeat."""
+        with self.lock:
+            if node_id in self.node_ids:
+                self.node_ids[node_id] = (time.time() + ping_interval, ping_interval)
+                return True
+        return False

flwr/server/superlink/state/sqlite_state.py

@@ -18,7 +18,7 @@
 import os
 import re
 import sqlite3
-from datetime import datetime, timedelta
+import time
 from logging import DEBUG, ERROR
 from typing import Any, Dict, List, Optional, Set, Tuple, Union, cast
 from uuid import UUID, uuid4
@@ -33,10 +33,16 @@ from .state import State
 
 SQL_CREATE_TABLE_NODE = """
 CREATE TABLE IF NOT EXISTS node(
-    node_id INTEGER UNIQUE
+    node_id         INTEGER UNIQUE,
+    online_until    REAL,
+    ping_interval   REAL
 );
 """
 
+SQL_CREATE_INDEX_ONLINE_UNTIL = """
+CREATE INDEX IF NOT EXISTS idx_online_until ON node (online_until);
+"""
+
 SQL_CREATE_TABLE_RUN = """
 CREATE TABLE IF NOT EXISTS run(
     run_id INTEGER UNIQUE
@@ -52,9 +58,10 @@ CREATE TABLE IF NOT EXISTS task_ins(
     producer_node_id        INTEGER,
     consumer_anonymous      BOOLEAN,
     consumer_node_id        INTEGER,
-    created_at              TEXT,
+    created_at              REAL,
     delivered_at            TEXT,
-    ttl                     TEXT,
+    pushed_at               REAL,
+    ttl                     REAL,
     ancestry                TEXT,
     task_type               TEXT,
     recordset               BLOB,
@@ -72,9 +79,10 @@ CREATE TABLE IF NOT EXISTS task_res(
     producer_node_id        INTEGER,
     consumer_anonymous      BOOLEAN,
     consumer_node_id        INTEGER,
-    created_at              TEXT,
+    created_at              REAL,
     delivered_at            TEXT,
-    ttl                     TEXT,
+    pushed_at               REAL,
+    ttl                     REAL,
     ancestry                TEXT,
     task_type               TEXT,
     recordset               BLOB,
@@ -82,7 +90,7 @@ CREATE TABLE IF NOT EXISTS task_res(
 );
 """
 
-DictOrTuple = Union[Tuple[Any], Dict[str, Any]]
+DictOrTuple = Union[Tuple[Any, ...], Dict[str, Any]]
 
 
 class SqliteState(State):
@@ -123,6 +131,7 @@ class SqliteState(State):
         cur.execute(SQL_CREATE_TABLE_TASK_INS)
         cur.execute(SQL_CREATE_TABLE_TASK_RES)
         cur.execute(SQL_CREATE_TABLE_NODE)
+        cur.execute(SQL_CREATE_INDEX_ONLINE_UNTIL)
         res = cur.execute("SELECT name FROM sqlite_schema;")
 
         return res.fetchall()
@@ -185,15 +194,11 @@ class SqliteState(State):
             log(ERROR, errors)
             return None
 
-        # Create task_id, created_at and ttl
+        # Create task_id
         task_id = uuid4()
-        created_at: datetime = now()
-        ttl: datetime = created_at + timedelta(hours=24)
 
         # Store TaskIns
         task_ins.task_id = str(task_id)
-        task_ins.task.created_at = created_at.isoformat()
-        task_ins.task.ttl = ttl.isoformat()
         data = (task_ins_to_dict(task_ins),)
         columns = ", ".join([f":{key}" for key in data[0]])
         query = f"INSERT INTO task_ins VALUES({columns});"
@@ -320,15 +325,11 @@ class SqliteState(State):
             log(ERROR, errors)
             return None
 
-        # Create task_id, created_at and ttl
+        # Create task_id
         task_id = uuid4()
-        created_at: datetime = now()
-        ttl: datetime = created_at + timedelta(hours=24)
 
         # Store TaskIns
         task_res.task_id = str(task_id)
-        task_res.task.created_at = created_at.isoformat()
-        task_res.task.ttl = ttl.isoformat()
         data = (task_res_to_dict(task_res),)
         columns = ", ".join([f":{key}" for key in data[0]])
         query = f"INSERT INTO task_res VALUES({columns});"
@@ -472,9 +473,14 @@ class SqliteState(State):
         # Sample a random int64 as node_id
         node_id: int = int.from_bytes(os.urandom(8), "little", signed=True)
 
-        query = "INSERT INTO node VALUES(:node_id);"
+        query = (
+            "INSERT INTO node (node_id, online_until, ping_interval) VALUES (?, ?, ?)"
+        )
+
         try:
-            self.query(query, {"node_id": node_id})
+            # Default ping interval is 30s
+            # TODO: change 1e9 to 30s  # pylint: disable=W0511
+            self.query(query, (node_id, time.time() + 1e9, 1e9))
         except sqlite3.IntegrityError:
             log(ERROR, "Unexpected node registration failure.")
             return 0
@@ -499,8 +505,8 @@ class SqliteState(State):
             return set()
 
         # Get nodes
-        query = "SELECT * FROM node;"
-        rows = self.query(query)
+        query = "SELECT node_id FROM node WHERE online_until > ?;"
+        rows = self.query(query, (time.time(),))
         result: Set[int] = {row["node_id"] for row in rows}
         return result
 
@@ -519,6 +525,17 @@ class SqliteState(State):
         log(ERROR, "Unexpected run creation failure.")
         return 0
 
+    def acknowledge_ping(self, node_id: int, ping_interval: float) -> bool:
+        """Acknowledge a ping received from a node, serving as a heartbeat."""
+        # Update `online_until` and `ping_interval` for the given `node_id`
+        query = "UPDATE node SET online_until = ?, ping_interval = ? WHERE node_id = ?;"
+        try:
+            self.query(query, (time.time() + ping_interval, ping_interval, node_id))
+            return True
+        except sqlite3.IntegrityError:
+            log(ERROR, "`node_id` does not exist.")
+            return False
+
 
 def dict_factory(
     cursor: sqlite3.Cursor,
@@ -544,6 +561,7 @@ def task_ins_to_dict(task_msg: TaskIns) -> Dict[str, Any]:
         "consumer_node_id": task_msg.task.consumer.node_id,
         "created_at": task_msg.task.created_at,
         "delivered_at": task_msg.task.delivered_at,
+        "pushed_at": task_msg.task.pushed_at,
         "ttl": task_msg.task.ttl,
         "ancestry": ",".join(task_msg.task.ancestry),
         "task_type": task_msg.task.task_type,
@@ -564,6 +582,7 @@ def task_res_to_dict(task_msg: TaskRes) -> Dict[str, Any]:
         "consumer_node_id": task_msg.task.consumer.node_id,
         "created_at": task_msg.task.created_at,
         "delivered_at": task_msg.task.delivered_at,
+        "pushed_at": task_msg.task.pushed_at,
         "ttl": task_msg.task.ttl,
         "ancestry": ",".join(task_msg.task.ancestry),
         "task_type": task_msg.task.task_type,
@@ -592,6 +611,7 @@ def dict_to_task_ins(task_dict: Dict[str, Any]) -> TaskIns:
             ),
             created_at=task_dict["created_at"],
             delivered_at=task_dict["delivered_at"],
+            pushed_at=task_dict["pushed_at"],
             ttl=task_dict["ttl"],
             ancestry=task_dict["ancestry"].split(","),
             task_type=task_dict["task_type"],
@@ -621,6 +641,7 @@ def dict_to_task_res(task_dict: Dict[str, Any]) -> TaskRes:
             ),
             created_at=task_dict["created_at"],
             delivered_at=task_dict["delivered_at"],
+            pushed_at=task_dict["pushed_at"],
             ttl=task_dict["ttl"],
             ancestry=task_dict["ancestry"].split(","),
             task_type=task_dict["task_type"],

flwr/server/superlink/state/state.py

@@ -152,3 +152,22 @@ class State(abc.ABC):
     @abc.abstractmethod
     def create_run(self) -> int:
         """Create one run."""
+
+    @abc.abstractmethod
+    def acknowledge_ping(self, node_id: int, ping_interval: float) -> bool:
+        """Acknowledge a ping received from a node, serving as a heartbeat.
+
+        Parameters
+        ----------
+        node_id : int
+            The `node_id` from which the ping was received.
+        ping_interval : float
+            The interval (in seconds) from the current timestamp within which the next
+            ping from this node must be received. This acts as a hard deadline to ensure
+            an accurate assessment of the node's availability.
+
+        Returns
+        -------
+        is_acknowledged : bool
+            True if the ping is successfully acknowledged; otherwise, False.
+        """

flwr/server/utils/validator.py

@@ -31,13 +31,21 @@ def validate_task_ins_or_res(tasks_ins_res: Union[TaskIns, TaskRes]) -> List[str
     if not tasks_ins_res.HasField("task"):
         validation_errors.append("`task` does not set field `task`")
 
-    # Created/delivered/TTL
-    if tasks_ins_res.task.created_at != "":
-        validation_errors.append("`created_at` must be an empty str")
+    # Created/delivered/TTL/Pushed
+    if (
+        tasks_ins_res.task.created_at < 1711497600.0
+    ):  # unix timestamp of 27 March 2024 00h:00m:00s UTC
+        validation_errors.append(
+            "`created_at` must be a float that records the unix timestamp "
+            "in seconds when the message was created."
+        )
     if tasks_ins_res.task.delivered_at != "":
         validation_errors.append("`delivered_at` must be an empty str")
-    if tasks_ins_res.task.ttl != "":
-        validation_errors.append("`ttl` must be an empty str")
+    if tasks_ins_res.task.ttl <= 0:
+        validation_errors.append("`ttl` must be higher than zero")
+    if tasks_ins_res.task.pushed_at < 1711497600.0:
+        # unix timestamp of 27 March 2024 00h:00m:00s UTC
+        validation_errors.append("`pushed_at` is not a recent timestamp")
 
     # TaskIns specific
     if isinstance(tasks_ins_res, TaskIns):
@@ -66,8 +74,11 @@ def validate_task_ins_or_res(tasks_ins_res: Union[TaskIns, TaskRes]) -> List[str
         # Content check
         if tasks_ins_res.task.task_type == "":
             validation_errors.append("`task_type` MUST be set")
-        if not tasks_ins_res.task.HasField("recordset"):
-            validation_errors.append("`recordset` MUST be set")
+        if not (
+            tasks_ins_res.task.HasField("recordset")
+            ^ tasks_ins_res.task.HasField("error")
+        ):
+            validation_errors.append("Either `recordset` or `error` MUST be set")
 
         # Ancestors
         if len(tasks_ins_res.task.ancestry) != 0:
@@ -106,8 +117,11 @@ def validate_task_ins_or_res(tasks_ins_res: Union[TaskIns, TaskRes]) -> List[str
         # Content check
         if tasks_ins_res.task.task_type == "":
             validation_errors.append("`task_type` MUST be set")
-        if not tasks_ins_res.task.HasField("recordset"):
-            validation_errors.append("`recordset` MUST be set")
+        if not (
+            tasks_ins_res.task.HasField("recordset")
+            ^ tasks_ins_res.task.HasField("error")
+        ):
+            validation_errors.append("Either `recordset` or `error` MUST be set")
 
         # Ancestors
         if len(tasks_ins_res.task.ancestry) == 0:

flwr/server/workflow/default_workflows.py

@@ -21,7 +21,7 @@ from logging import INFO
 from typing import Optional, cast
 
 import flwr.common.recordset_compat as compat
-from flwr.common import ConfigsRecord, Context, GetParametersIns, log
+from flwr.common import DEFAULT_TTL, ConfigsRecord, Context, GetParametersIns, log
 from flwr.common.constant import MessageType, MessageTypeLegacy
 
 from ..compat.app_utils import start_update_client_manager_thread
@@ -127,7 +127,7 @@ def default_init_params_workflow(driver: Driver, context: Context) -> None:
                     message_type=MessageTypeLegacy.GET_PARAMETERS,
                     dst_node_id=random_client.node_id,
                     group_id="0",
-                    ttl="",
+                    ttl=DEFAULT_TTL,
                 )
             ]
         )
@@ -226,7 +226,7 @@ def default_fit_workflow(  # pylint: disable=R0914
             message_type=MessageType.TRAIN,
             dst_node_id=proxy.node_id,
             group_id=str(current_round),
-            ttl="",
+            ttl=DEFAULT_TTL,
         )
         for proxy, fitins in client_instructions
     ]
@@ -306,7 +306,7 @@ def default_evaluate_workflow(driver: Driver, context: Context) -> None:
             message_type=MessageType.EVALUATE,
             dst_node_id=proxy.node_id,
             group_id=str(current_round),
-            ttl="",
+            ttl=DEFAULT_TTL,
         )
         for proxy, evalins in client_instructions
     ]

flwr/server/workflow/secure_aggregation/secaggplus_workflow.py

@@ -22,6 +22,7 @@ from typing import Dict, List, Optional, Set, Tuple, Union, cast
 
 import flwr.common.recordset_compat as compat
 from flwr.common import (
+    DEFAULT_TTL,
     ConfigsRecord,
     Context,
     FitRes,
@@ -373,7 +374,7 @@ class SecAggPlusWorkflow:
                 message_type=MessageType.TRAIN,
                 dst_node_id=nid,
                 group_id=str(cfg[WorkflowKey.CURRENT_ROUND]),
-                ttl="",
+                ttl=DEFAULT_TTL,
             )
 
         log(
@@ -421,7 +422,7 @@ class SecAggPlusWorkflow:
                 message_type=MessageType.TRAIN,
                 dst_node_id=nid,
                 group_id=str(cfg[WorkflowKey.CURRENT_ROUND]),
-                ttl="",
+                ttl=DEFAULT_TTL,
             )
 
         # Broadcast public keys to clients and receive secret key shares
@@ -492,7 +493,7 @@ class SecAggPlusWorkflow:
                 message_type=MessageType.TRAIN,
                 dst_node_id=nid,
                 group_id=str(cfg[WorkflowKey.CURRENT_ROUND]),
-                ttl="",
+                ttl=DEFAULT_TTL,
             )
 
         log(
@@ -563,7 +564,7 @@ class SecAggPlusWorkflow:
                 message_type=MessageType.TRAIN,
                 dst_node_id=nid,
                 group_id=str(current_round),
-                ttl="",
+                ttl=DEFAULT_TTL,
             )
 
         log(

flwr/simulation/ray_transport/ray_client_proxy.py

@@ -23,7 +23,7 @@ from flwr import common
 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 import DEFAULT_TTL, Message, Metadata, RecordSet
 from flwr.common.constant import MessageType, MessageTypeLegacy
 from flwr.common.logger import log
 from flwr.common.recordset_compat import (
@@ -105,7 +105,7 @@ class RayActorClientProxy(ClientProxy):
                 src_node_id=0,
                 dst_node_id=int(self.cid),
                 reply_to_message="",
-                ttl=str(timeout) if timeout else "",
+                ttl=timeout if timeout else DEFAULT_TTL,
                 message_type=message_type,
                 partition_id=int(self.cid),
             ),

{flwr_nightly-1.8.0.dev20240315.dist-info → flwr_nightly-1.8.0.dev20240327.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: flwr-nightly
-Version: 1.8.0.dev20240315
+Version: 1.8.0.dev20240327
 Summary: Flower: A Friendly Federated Learning Framework
 Home-page: https://flower.ai
 License: Apache-2.0