Add wait_for_device info and timeout

nvflare/edge/assessors/buff_device_manager.py

@@ -89,7 +89,10 @@ def remove_devices_from_used(self, devices: Set[str], fl_ctx) -> None:
             self.used_devices.pop(device_id, None)
 
     def has_enough_devices(self, fl_ctx) -> bool:
-        return len(self.available_devices) >= self.device_selection_size
+        num_holes = self.device_selection_size - len(self.current_selection)
+        usable_devices = set(self.available_devices.keys()) - set(self.used_devices.keys())
+        num_usable_devices = len(usable_devices)
+        return num_usable_devices >= num_holes
 
     def should_fill_selection(self, fl_ctx) -> bool:
         num_holes = self.device_selection_size - len(self.current_selection)

nvflare/edge/assessors/buff_model_manager.py

@@ -154,19 +154,15 @@ def process_updates(self, model_updates: Dict[int, ModelUpdate], fl_ctx: FLConte
                 self.log_error(fl_ctx, f"bad child update version {model_version}: no update data")
                 continue
 
-            if self.current_model_version - model_version > self.max_model_history:
+            if self.current_model_version - model_version >= self.max_model_history:
                 # this version is too old
-                self.log_info(
+                self.log_warning(
                     fl_ctx,
-                    f"dropped child update version {model_version}. Current version {self.current_model_version}",
+                    f"dropped child update version {model_version}. Current version {self.current_model_version}. Max history {self.max_model_history}",
                 )
                 continue
 
             model_state = self.updates.get(model_version)
-            if not model_state:
-                self.log_error(fl_ctx, f"No model state for version {model_version}")
-                continue
-
             accepted = model_state.accept(model_update, fl_ctx)
             self.log_info(
                 fl_ctx,

nvflare/edge/assessors/model_update.py

@@ -14,6 +14,7 @@
 
 import threading
 import time
+from concurrent.futures import ThreadPoolExecutor
 from typing import Optional
 
 from nvflare.apis.event_type import EventType
@@ -33,6 +34,7 @@ def __init__(
         model_manager_id,
         device_manager_id,
         max_model_version,
+        device_wait_timeout: float = 30.0,
     ):
         """Initialize the ModelUpdateAssessor.
         Enable both asynchronous and synchronous model updates from clients.
@@ -44,6 +46,7 @@ def __init__(
             model_manager_id (str): ID of the model manager component.
             device_manager_id (str): ID of the device manager component.
             max_model_version (int): Maximum model version to stop the workflow.
+            device_wait_timeout (float): Timeout in seconds for waiting for sufficient devices. Default is 30 seconds.
         """
         Assessor.__init__(self)
         self.persistor_id = persistor_id
@@ -53,10 +56,132 @@ def __init__(
         self.model_manager = None
         self.device_manager = None
         self.max_model_version = max_model_version
+        self.device_wait_timeout = device_wait_timeout
         self.update_lock = threading.Lock()
-        self.start_time = None
+        self.device_wait_start_time = None
+        self.should_stop_job = False
+        self.timeout_future = None
+        self.thread_pool = ThreadPoolExecutor(max_workers=1, thread_name_prefix="DeviceTimeout")
         self.register_event_handler(EventType.START_RUN, self._handle_start_run)
 
+    def _check_device_timeout(self, fl_ctx: FLContext) -> bool:
+        """Check if device wait timeout has been exceeded.
+
+        Args:
+            fl_ctx: FL context
+
+        Returns:
+            bool: True if timeout exceeded, False otherwise
+        """
+        if self.device_wait_start_time is None:
+            return False
+
+        if time.time() - self.device_wait_start_time > self.device_wait_timeout:
+            return True
+        return False
+
+    def _log_device_wait_status(self, fl_ctx: FLContext, message_prefix: str = ""):
+        """Log current device wait status with countdown information."""
+        if self.device_wait_start_time is not None:
+            current_time = time.time()
+            elapsed = current_time - self.device_wait_start_time
+
+            # Only log if we haven't logged recently (rate limiting)
+            if not hasattr(self, "_last_status_log_time"):
+                self._last_status_log_time = 0
+
+            # Adaptive logging frequency based on urgency
+            remaining_time = self.device_wait_timeout - elapsed
+            if remaining_time > 0:
+                # Determine logging interval based on remaining time
+                if remaining_time > 60:  # More than 1 minute: log every 30 seconds
+                    log_interval = 30.0
+                elif remaining_time > 30:  # 30 seconds to 1 minute: log every 15 seconds
+                    log_interval = 15.0
+                elif remaining_time > 10:  # 10 to 30 seconds: log every 5 seconds
+                    log_interval = 5.0
+                else:  # Final 10 seconds: log every 2 seconds
+                    log_interval = 2.0
+
+                # Check if enough time has passed since last log
+                if current_time - self._last_status_log_time >= log_interval:
+                    usable_devices = set(self.device_manager.available_devices.keys()) - set(
+                        self.device_manager.used_devices.keys()
+                    )
+                    self.log_info(
+                        fl_ctx,
+                        f"{message_prefix}Device wait status: "
+                        f"Total devices: {len(self.device_manager.available_devices)}, "
+                        f"usable: {len(usable_devices)}, "
+                        f"expected: {self.device_manager.device_selection_size}. "
+                        f"Timeout in {remaining_time:.1f} seconds.",
+                    )
+                    self._last_status_log_time = current_time
+
+    def _start_timeout_tracking(self, fl_ctx: FLContext):
+        """Start independent timeout tracking using thread pool."""
+        if self.timeout_future is not None and not self.timeout_future.done():
+            # Cancel existing timeout if still running
+            self.timeout_future.cancel()
+
+        # Submit new timeout task to thread pool
+        self.timeout_future = self.thread_pool.submit(self._timeout_tracker, fl_ctx)
+        self.log_debug(fl_ctx, f"Started device wait timeout tracking for {self.device_wait_timeout}s")
+
+    def _stop_timeout_tracking(self):
+        """Stop timeout tracking."""
+        if self.timeout_future is not None and not self.timeout_future.done():
+            self.timeout_future.cancel()
+            self.timeout_future = None
+
+    def _timeout_tracker(self, fl_ctx: FLContext):
+        """Independent timeout tracker that runs in thread pool."""
+        try:
+            # Periodic logging during countdown to keep users informed
+            check_interval = 10.0  # Log every 10 seconds
+            elapsed = 0
+
+            while (
+                self.device_wait_start_time is not None
+                and not self.should_stop_job
+                and elapsed < self.device_wait_timeout
+            ):
+
+                time.sleep(min(check_interval, self.device_wait_timeout - elapsed))
+                elapsed += check_interval
+
+                # Check if we should stop monitoring
+                if self.device_wait_start_time is None or self.should_stop_job:
+                    return
+
+                # Log periodic status update
+                if elapsed < self.device_wait_timeout:
+                    remaining = self.device_wait_timeout - elapsed
+                    self.log_info(fl_ctx, f"Device wait countdown: {remaining:.1f} seconds remaining")
+
+            # Check if we're still waiting for devices and timeout exceeded
+            with self.update_lock:
+                if (
+                    self.device_wait_start_time is not None
+                    and not self.should_stop_job
+                    and self._check_device_timeout(fl_ctx)
+                ):
+
+                    # Timeout exceeded, set the stop flag
+                    self.should_stop_job = True
+                    self.log_error(
+                        fl_ctx,
+                        f"Device wait timeout ({self.device_wait_timeout}s) exceeded. "
+                        f"Setting stop job flag to terminate workflow.",
+                    )
+
+        except Exception as e:
+            # Log error but don't crash the thread
+            self.log_error(fl_ctx, f"Error in timeout tracker: {e}")
+        finally:
+            # Clean up
+            self.timeout_future = None
+
     def _handle_start_run(self, event_type: str, fl_ctx: FLContext):
         engine = fl_ctx.get_engine()
 
@@ -96,7 +221,6 @@ def _handle_start_run(self, event_type: str, fl_ctx: FLContext):
             return
 
     def start_task(self, fl_ctx: FLContext) -> Shareable:
-        self.start_time = time.time()
         # empty base state to start with
         base_state = BaseState(
             model_version=0,
@@ -116,6 +240,38 @@ def _do_child_update(self, update: Shareable, fl_ctx: FLContext) -> (bool, Optio
         # Update available devices
         if report.available_devices:
             self.device_manager.update_available_devices(report.available_devices, fl_ctx)
+            # Reset wait timer if we now have enough devices
+            if self.device_wait_start_time is not None and self.device_manager.has_enough_devices(fl_ctx):
+                self.device_wait_start_time = None
+                self.should_stop_job = False  # Reset stop job flag
+                # Stop timeout tracking since we have enough devices
+                self._stop_timeout_tracking()
+                self.log_info(fl_ctx, "Sufficient devices now available, resetting wait timer and stop job flag")
+
+        # Check for device wait timeout if we are waiting for devices
+        if self.device_wait_start_time is not None:
+            if self.should_stop_job:
+                # Timeout exceeded, prepare an empty reply
+                usable_devices = set(self.device_manager.available_devices.keys()) - set(
+                    self.device_manager.used_devices.keys()
+                )
+                self.log_error(
+                    fl_ctx,
+                    f"Device wait timeout ({self.device_wait_timeout}s) exceeded. "
+                    f"Total devices: {len(self.device_manager.available_devices)}, usable: {len(usable_devices)}, expected: {self.device_manager.device_selection_size}. "
+                    f"Device_reuse flag is set to: {self.device_manager.device_reuse}. "
+                    "Not enough devices joining, please adjust the server params. Stopping the job.",
+                )
+                reply = StateUpdateReply(
+                    model_version=0,
+                    model=None,
+                    device_selection_version=self.device_manager.current_selection_version,
+                    device_selection=self.device_manager.get_selection(fl_ctx),
+                )
+                return False, reply.to_shareable()
+            else:
+                # Log current wait status for user information (rate limited)
+                self._log_device_wait_status(fl_ctx, "Waiting for devices: ")
 
         accepted = True
         if report.model_updates:
@@ -133,21 +289,27 @@ def _do_child_update(self, update: Shareable, fl_ctx: FLContext) -> (bool, Optio
                     if self.device_manager.device_reuse:
                         self.device_manager.remove_devices_from_used(set(model_update.devices.keys()), fl_ctx)
 
-            # Handle device selection
-            if self.device_manager.should_fill_selection(fl_ctx):
-                self.device_manager.fill_selection(self.model_manager.current_model_version, fl_ctx)
-
         else:
             self.log_debug(fl_ctx, "no model updates")
 
-        # Handle initial model generation
-        if self.model_manager.current_model_version == 0:
+        # Handle device selection
+        if self.device_manager.should_fill_selection(fl_ctx):
+            # check if we have enough devices to fill selection
             if self.device_manager.has_enough_devices(fl_ctx):
-                self.log_info(
-                    fl_ctx, f"got {len(self.device_manager.available_devices)} devices - generate initial model"
-                )
-                self.model_manager.generate_new_model(fl_ctx)
+                if self.model_manager.current_model_version == 0:
+                    self.log_info(fl_ctx, "Generate initial model and fill selection")
+                    self.model_manager.generate_new_model(fl_ctx)
                 self.device_manager.fill_selection(self.model_manager.current_model_version, fl_ctx)
+                # Reset wait timer since we have enough devices
+                self.device_wait_start_time = None
+                # Stop timeout tracking since we have enough devices
+                self._stop_timeout_tracking()
+            else:
+                # Start or continue wait timer since we don't have enough devices
+                if self.device_wait_start_time is None:
+                    self.device_wait_start_time = time.time()
+                    # Start independent timeout tracking
+                    self._start_timeout_tracking(fl_ctx)
 
         # Prepare reply
         model = None
@@ -163,9 +325,21 @@ def _do_child_update(self, update: Shareable, fl_ctx: FLContext) -> (bool, Optio
         return accepted, reply.to_shareable()
 
     def assess(self, fl_ctx: FLContext) -> Assessment:
-        if self.model_manager.current_model_version >= self.max_model_version:
+        if self.should_stop_job:
+            # Stop timeout tracking before ending the job
+            self._stop_timeout_tracking()
+            self.log_error(fl_ctx, "Job stopped due to insufficient devices joining within timeout period")
+            return Assessment.WORKFLOW_DONE
+        elif self.model_manager.current_model_version >= self.max_model_version:
+            # Stop timeout tracking before ending the job
+            self._stop_timeout_tracking()
             model_version = self.model_manager.current_model_version
             self.log_info(fl_ctx, f"Max model version {self.max_model_version} reached: {model_version=}")
             return Assessment.WORKFLOW_DONE
         else:
             return Assessment.CONTINUE
+
+    def __del__(self):
+        """Cleanup thread pool on destruction."""
+        if hasattr(self, "thread_pool"):
+            self.thread_pool.shutdown(wait=False)