experimaestro 2.0.0a8__py3-none-any.whl → 2.0.0b4__py3-none-any.whl

experimaestro/scheduler/base.py

@@ -1,15 +1,16 @@
-import logging
 import threading
 import time
 from typing import (
     Optional,
     Set,
+    ClassVar,
+    TYPE_CHECKING,
 )
 import asyncio
 from typing import Dict
 
 from experimaestro.scheduler import experiment
-from experimaestro.scheduler.jobs import Job, JobState
+from experimaestro.scheduler.jobs import Job, JobState, JobError
 from experimaestro.scheduler.services import Service
 
 
@@ -17,6 +18,11 @@ from experimaestro.utils import logger
 from experimaestro.utils.asyncio import asyncThreadcheck
 import concurrent.futures
 
+if TYPE_CHECKING:
+    from experimaestro.server import Server
+    from experimaestro.settings import ServerSettings
+    from experimaestro.scheduler.workspace import Workspace
+
 
 class Listener:
     def job_submitted(self, job):
@@ -31,18 +37,24 @@ class Listener:
 
 
 class Scheduler(threading.Thread):
-    """A job scheduler
+    """A job scheduler (singleton)
 
-    The scheduler is based on asyncio for easy concurrency handling
+    The scheduler is based on asyncio for easy concurrency handling.
+    This is a singleton - only one scheduler instance exists per process.
     """
 
-    def __init__(self, xp: "experiment", name: str):
+    _instance: ClassVar[Optional["Scheduler"]] = None
+    _lock: ClassVar[threading.Lock] = threading.Lock()
+
+    def __init__(self, name: str = "Global"):
         super().__init__(name=f"Scheduler ({name})", daemon=True)
         self._ready = threading.Event()
 
-        # Name of the experiment
+        # Name of the scheduler
         self.name = name
-        self.xp = xp
+
+        # Track experiments (simple dict for now)
+        self.experiments: Dict[str, "experiment"] = {}
 
         # Exit mode activated
         self.exitmode = False
@@ -53,16 +65,101 @@ class Scheduler(threading.Thread):
         # List of jobs
         self.waitingjobs: Set[Job] = set()
 
-        # Listeners
-        self.listeners: Set[Listener] = set()
+        # Listeners with thread-safe access
+        self._listeners: Set[Listener] = set()
+        self._listeners_lock = threading.Lock()
+
+        # Notification thread pool (single worker to serialize notifications)
+        self._notification_executor = concurrent.futures.ThreadPoolExecutor(
+            max_workers=1, thread_name_prefix="NotificationWorker"
+        )
+
+        # Server (managed by scheduler)
+        self.server: Optional["Server"] = None
 
     @staticmethod
-    def create(xp: "experiment", name: str):
-        instance = Scheduler(xp, name)
+    def has_instance() -> bool:
+        """Check if a scheduler instance exists without creating one"""
+        return Scheduler._instance is not None
+
+    @staticmethod
+    def instance() -> "Scheduler":
+        """Get or create the global scheduler instance"""
+        if Scheduler._instance is None:
+            with Scheduler._lock:
+                if Scheduler._instance is None:
+                    Scheduler._instance = Scheduler._create()
+        return Scheduler._instance
+
+    @staticmethod
+    def _create(name: str = "Global"):
+        """Internal method to create and start scheduler"""
+        instance = Scheduler(name)
         instance.start()
         instance._ready.wait()
         return instance
 
+    @staticmethod
+    def create(xp: "experiment" = None, name: str = "Global"):
+        """Create or get the scheduler instance
+
+        Args:
+            xp: (Deprecated) Experiment reference, ignored
+            name: Name for the scheduler (only used on first creation)
+
+        Returns:
+            The global scheduler instance
+        """
+        return Scheduler.instance()
+
+    def register_experiment(self, xp: "experiment"):
+        """Register an experiment with the scheduler"""
+        # Use experiment name as key for now
+        key = xp.workdir.name
+        self.experiments[key] = xp
+
+        logger.debug("Registered experiment %s with scheduler", key)
+
+    def unregister_experiment(self, xp: "experiment"):
+        """Unregister an experiment from the scheduler"""
+        key = xp.workdir.name
+        if key in self.experiments:
+            del self.experiments[key]
+            logger.debug("Unregistered experiment %s from scheduler", key)
+
+    def start_server(
+        self, settings: "ServerSettings" = None, workspace: "Workspace" = None
+    ):
+        """Start the notification server (if not already running)
+
+        Args:
+            settings: Server settings
+            workspace: Workspace instance (required to get workspace path)
+        """
+        if self.server is None:
+            from experimaestro.server import Server
+            from experimaestro.scheduler.state_provider import WorkspaceStateProvider
+
+            if workspace is None:
+                raise ValueError("workspace parameter is required to start server")
+
+            # Get the workspace state provider singleton
+            state_provider = WorkspaceStateProvider.get_instance(
+                workspace.path, read_only=False, sync_on_start=False
+            )
+
+            self.server = Server.instance(settings, state_provider)
+            self.server.start()
+            logger.info("Server started by scheduler")
+        else:
+            logger.debug("Server already running")
+
+    def stop_server(self):
+        """Stop the notification server"""
+        if self.server is not None:
+            self.server.stop()
+            logger.info("Server stopped by scheduler")
+
     def run(self):
         """Run the event loop forever"""
         logger.debug("Starting event loop thread")
@@ -72,6 +169,10 @@ class Scheduler(threading.Thread):
         # Set loop-dependent variables
         self.exitCondition = asyncio.Condition()
         self.dependencyLock = asyncio.Lock()
+
+        # Note: State provider removed - now managed at workspace level
+        # Each experiment has its own workspace with database
+
         self._ready.set()
         self.loop.run_forever()
 
@@ -84,10 +185,17 @@ class Scheduler(threading.Thread):
             logger.warning("Scheduler already started")
 
     def addlistener(self, listener: Listener):
-        self.listeners.add(listener)
+        with self._listeners_lock:
+            self._listeners.add(listener)
 
     def removelistener(self, listener: Listener):
-        self.listeners.remove(listener)
+        with self._listeners_lock:
+            self._listeners.discard(listener)
+
+    def clear_listeners(self):
+        """Clear all listeners (for testing purposes)"""
+        with self._listeners_lock:
+            self._listeners.clear()
 
     def getJobState(self, job: Job) -> "concurrent.futures.Future[JobState]":
         # Check if the job belongs to this scheduler
@@ -104,17 +212,25 @@ class Scheduler(threading.Thread):
 
     def submit(self, job: Job) -> Optional[Job]:
         # Wait for the future containing the submitted job
-        logger.debug("Registering the job %s within the scheduler", job)
+        logger.debug("Submit job %s to the scheduler", job)
         otherFuture = asyncio.run_coroutine_threadsafe(
             self.aio_registerJob(job), self.loop
         )
         other = otherFuture.result()
         logger.debug("Job already submitted" if other else "First submission")
-        if other:
-            return other
+
+        # Only returns if job was already submitted and doesn't need reprocessing
+        if other is not None:
+            # If state is WAITING, it was just reset for resubmission and needs processing
+            # If state is RUNNING or finished (DONE), no need to reprocess
+            if other.state != JobState.WAITING:
+                return other
+            # Use 'other' for resubmission since it has the correct experiments list
+            job = other
 
         job._future = asyncio.run_coroutine_threadsafe(self.aio_submit(job), self.loop)
-        return None
+
+        return other
 
     def prepare(self, job: Job):
         """Prepares the job for running"""
@@ -129,49 +245,99 @@ class Scheduler(threading.Thread):
 
         if self.exitmode:
             logger.warning("Exit mode: not submitting")
+            return
 
-        elif job.identifier in self.jobs:
+        # Job was already submitted
+        if job.identifier in self.jobs:
             other = self.jobs[job.identifier]
             assert job.type == other.type
-            if other.state == JobState.ERROR:
+
+            # Add current experiment to the existing job's experiments list
+            xp = experiment.current()
+            xp.add_job(other)
+
+            # Copy watched outputs from new job to existing job
+            # This ensures new callbacks are registered even for resubmitted jobs
+            other.watched_outputs.extend(job.watched_outputs)
+
+            if other.state.is_error():
                 logger.info("Re-submitting job")
+                # Clean up old process info so it will be re-started
+                other._process = None
+                if other.pidpath.is_file():
+                    other.pidpath.unlink()
+                # Use set_state to handle experiment statistics updates
+                other.set_state(JobState.WAITING)
+                self.notify_job_state(other)  # Notify listeners of re-submit
             else:
                 logger.warning("Job %s already submitted", job.identifier)
-                return other
 
-        else:
-            # Register this job
-            self.xp.unfinishedJobs += 1
-            self.jobs[job.identifier] = job
+            # Returns the previous job
+            return other
+
+        # Register this job
+        xp = experiment.current()
+        self.jobs[job.identifier] = job
+        # Set submittime now so that add_job can record it in the database
+        # (aio_submit may update this later for re-submitted jobs)
+        job.submittime = time.time()
+        xp.add_job(job)
+
+        # Set up dependencies
+        for dependency in job.dependencies:
+            dependency.target = job
+            dependency.origin.dependents.add(dependency)
 
         return None
 
+    def _notify_listeners(self, notification_func, job: Job):
+        """Execute notification in thread pool with error isolation.
+
+        This runs notifications in a dedicated thread pool to avoid blocking
+        the scheduler and to isolate errors from affecting other listeners.
+        """
+
+        def _do_notify():
+            # Get a snapshot of listeners with the lock
+            with self._listeners_lock:
+                listeners_snapshot = list(self._listeners)
+
+            for listener in listeners_snapshot:
+                try:
+                    notification_func(listener, job)
+                except Exception:
+                    logger.exception("Got an error with listener %s", listener)
+
+        self._notification_executor.submit(_do_notify)
+
     def notify_job_submitted(self, job: Job):
         """Notify the listeners that a job has been submitted"""
-        for listener in self.listeners:
-            try:
-                listener.job_submitted(job)
-            except Exception:
-                logger.exception("Got an error with listener %s", listener)
+        self._notify_listeners(lambda lst, j: lst.job_submitted(j), job)
 
     def notify_job_state(self, job: Job):
         """Notify the listeners that a job has changed state"""
-        for listener in self.listeners:
-            try:
-                listener.job_state(job)
-            except Exception:
-                logger.exception("Got an error with listener %s", listener)
+        self._notify_listeners(lambda lst, j: lst.job_state(j), job)
 
-    async def aio_submit(self, job: Job) -> JobState:  # noqa: C901
+    def notify_service_add(self, service: Service):
+        """Notify the listeners that a service has been added"""
+        self._notify_listeners(lambda lst, s: lst.service_add(s), service)
+
+    async def aio_submit(self, job: Job) -> JobState:
         """Main scheduler function: submit a job, run it (if needed), and returns
         the status code
         """
+        from experimaestro.scheduler.jobs import JobStateError, JobFailureStatus
+
         logger.info("Submitting job %s", job)
-        job._readyEvent = asyncio.Event()
         job.submittime = time.time()
         job.scheduler = self
         self.waitingjobs.add(job)
 
+        # Register watched outputs now that the job has a scheduler
+        job.register_watched_outputs()
+
+        # Note: Job metadata will be written after directory is created in aio_start
+
         # Check that we don't have a completed job in
         # alternate directories
         for jobspath in experiment.current().alt_jobspaths:
@@ -185,126 +351,324 @@ class Scheduler(threading.Thread):
             path.unlink()
         path.symlink_to(job.path)
 
-        job.state = JobState.WAITING
-
+        job.set_state(JobState.WAITING)
         self.notify_job_submitted(job)
 
-        # Add dependencies, and add to blocking resources
-        if job.dependencies:
-            job.unsatisfied = len(job.dependencies)
-
-            for dependency in job.dependencies:
-                dependency.target = job
-                dependency.loop = self.loop
-                dependency.origin.dependents.add(dependency)
-                dependency.check()
-        else:
-            job._readyEvent.set()
-            job.state = JobState.READY
-
+        # Check if already done
         if job.donepath.exists():
-            job.state = JobState.DONE
+            job.set_state(JobState.DONE)
+            self.notify_job_state(job)  # Notify listeners of done state
 
         # Check if we have a running process
-        process = await job.aio_process()
-        if process is not None:
-            # Yep! First we notify the listeners
-            job.state = JobState.RUNNING
-            # Notify the listeners
-            self.notify_job_state(job)
-
-            # Adds to the listeners
-            if self.xp.server is not None:
-                job.add_notification_server(self.xp.server)
-
-            # And now, we wait...
-            logger.info("Got a process for job %s - waiting to complete", job)
-            code = await process.aio_code()
-            logger.info("Job %s completed with code %s", job, code)
-            job.state = JobState.DONE if code == 0 else JobState.ERROR
-
-        # Check if done
-        if job.donepath.exists():
-            job.state = JobState.DONE
-
-        # OK, not done; let's start the job for real
-        while not job.state.finished():
-            # Wait that the job is ready
-            await job._readyEvent.wait()
-            job._readyEvent.clear()
-
-            if job.state == JobState.READY:
-                try:
-                    state = await self.aio_start(job)
-                except Exception:
-                    logger.exception("Got an exception while starting the job")
-                    raise
+        if not job.state.finished():
+            process = await job.aio_process()
+            if process is not None:
+                # Notify listeners that job is running
+                job.set_state(JobState.RUNNING)
+                self.notify_job_state(job)
+
+                # Adds to the listeners
+                if self.server is not None:
+                    job.add_notification_server(self.server)
+
+                # And now, we wait...
+                logger.info("Got a process for job %s - waiting to complete", job)
+                code = await process.aio_code()
+                logger.info("Job %s completed with code %s", job, code)
+
+                # Record exit code if available
+                if code is not None:
+                    job.exit_code = code
+
+                # Read state from .done/.failed files (contains detailed failure reason)
+                state = JobState.from_path(job.path, job.name)
+
+                # If state is a generic FAILED error, let the process determine
+                # the state (it may detect launcher-specific failures like SLURM timeout)
+                if (
+                    state is not None
+                    and isinstance(state, JobStateError)
+                    and state.failure_reason == JobFailureStatus.FAILED
+                    and code is not None
+                ):
+                    process_state = process.get_job_state(code)
+                    if (
+                        isinstance(process_state, JobStateError)
+                        and process_state.failure_reason != JobFailureStatus.FAILED
+                    ):
+                        # Process detected a more specific failure reason
+                        state = process_state
 
                 if state is None:
-                    # State is None if this is not the main thread
-                    return JobState.ERROR
+                    if code is not None:
+                        # Fall back to process-specific state detection
+                        state = process.get_job_state(code)
+                    else:
+                        logger.error("No .done or .failed file found for job %s", job)
+                        state = JobState.ERROR
+                # Set endtime before set_state so database gets the timestamp
+                job.endtime = time.time()
+                job.set_state(state)
+                self.notify_job_state(job)  # Notify listeners of final state
+
+        # If not done or running, start the job
+        if not job.state.finished():
+            try:
+                state = await self.aio_start(job)
+                # Set endtime before set_state so database gets the timestamp
+                job.endtime = time.time()
+                job.set_state(state)
+            except Exception:
+                logger.exception("Got an exception while starting the job")
+                raise
 
-                job.state = state
+        # Job is finished - experiment statistics already updated by set_state
 
-        self.notify_job_state(job)
+        # Write final metadata with end time and final state
+        job.write_metadata()
 
-        # Job is finished
-        if job.state != JobState.DONE:
-            self.xp.failedJobs[job.identifier] = job
+        if job in self.waitingjobs:
+            self.waitingjobs.remove(job)
 
-        # Process all remaining tasks outputs
+        # Process all remaining task outputs BEFORE notifying exit condition
+        # This ensures taskOutputQueueSize is updated before wait() can check it,
+        # preventing a race where wait() sees both unfinishedJobs==0 and
+        # taskOutputQueueSize==0 before callbacks have been queued.
         await asyncThreadcheck("End of job processing", job.done_handler)
 
-        # Decrement the number of unfinished jobs and notify
-        self.xp.unfinishedJobs -= 1
+        # Now notify - wait() will see the correct taskOutputQueueSize
         async with self.exitCondition:
-            logging.debug("Updated number of unfinished jobs")
             self.exitCondition.notify_all()
 
-        job.endtime = time.time()
-        if job in self.waitingjobs:
-            self.waitingjobs.remove(job)
-
-        with job.dependents as dependents:
-            logger.info("Processing %d dependent jobs", len(dependents))
-            for dependency in dependents:
-                logger.debug("Checking dependency %s", dependency)
-                self.loop.call_soon(dependency.check)
-
         return job.state
 
-    async def aio_start(self, job: Job) -> Optional[JobState]:
-        """Start a job (scheduler coordination layer)
+    async def aio_start(self, job: Job) -> Optional[JobState]:  # noqa: C901
+        """Start a job with full job starting logic
 
-        This method serves as a coordination layer that delegates the actual
-        job starting logic to the job itself while handling scheduler-specific
-        concerns like state notifications and providing coordination context.
+        This method handles job locking, dependency acquisition, directory setup,
+        and job execution while using the scheduler's coordination lock to prevent
+        race conditions between multiple jobs.
 
         :param job: The job to start
         :return: JobState.WAITING if dependencies could not be locked, JobState.DONE
             if job completed successfully, JobState.ERROR if job failed during execution,
             or None (should not occur in normal operation)
-        :raises Exception: Various exceptions during scheduler coordination
+        :raises Exception: Various exceptions during job execution, dependency locking,
+            or process creation
         """
+        from experimaestro.scheduler.jobs import JobStateError
+        from experimaestro.locking import Locks, LockError
+        from experimaestro.scheduler.jobs import JobFailureStatus
 
         # Assert preconditions
         assert job.launcher is not None
 
-        try:
-            # Call job's start method with scheduler context
-            state = await job.aio_start(
-                sched_dependency_lock=self.dependencyLock,
-                notification_server=self.xp.server if self.xp else None,
+        # Restart loop for resumable tasks that timeout
+        while True:
+            logger.debug(
+                "Starting job %s with %d dependencies",
+                job,
+                len(job.dependencies),
             )
 
-            if state is None:
-                # Dependencies couldn't be locked, return WAITING state
-                return JobState.WAITING
+            # Separate static and dynamic dependencies
+            static_deps = [d for d in job.dependencies if not d.is_dynamic()]
+            dynamic_deps = [d for d in job.dependencies if d.is_dynamic()]
 
-            # Notify scheduler listeners of job state after successful start
+            # First, wait for all static dependencies (jobs) to complete
+            # These don't need the dependency lock as they can't change state
+            # Static dependency locks don't need to be added to locks list
+            logger.debug("Waiting for %d static dependencies", len(static_deps))
+            for dependency in static_deps:
+                logger.debug("Waiting for static dependency %s", dependency)
+                try:
+                    await dependency.aio_lock()
+                except RuntimeError as e:
+                    # Dependency failed - mark job as failed due to dependency
+                    logger.warning("Dependency failed: %s", e)
+                    return JobStateError(JobFailureStatus.DEPENDENCY)
+
+            # We first lock the job before proceeding
+            with Locks() as locks:
+                logger.debug("[starting] Locking job %s", job)
+                async with job.launcher.connector.lock(job.lockpath):
+                    logger.debug("[starting] Locked job %s", job)
+
+                    state = None
+                    try:
+                        # Now handle dynamic dependencies (tokens) with retry logic
+                        # CRITICAL: Only one task at a time can acquire dynamic dependencies
+                        # to prevent deadlocks (e.g., Task A holds Token1 waiting for Token2,
+                        # Task B holds Token2 waiting for Token1)
+                        if dynamic_deps:
+                            async with self.dependencyLock:
+                                logger.debug(
+                                    "Locking %d dynamic dependencies (tokens)",
+                                    len(dynamic_deps),
+                                )
+                                while True:
+                                    all_locked = True
+                                    for idx, dependency in enumerate(dynamic_deps):
+                                        try:
+                                            # Use timeout=0 for first dependency, 0.1s for subsequent
+                                            timeout = 0 if idx == 0 else 0.1
+                                            # Acquire the lock (this might block on IPC locks)
+                                            lock = await dependency.aio_lock(
+                                                timeout=timeout
+                                            )
+                                            locks.append(lock)
+                                        except LockError:
+                                            logger.info(
+                                                "Could not lock %s, retrying",
+                                                dependency,
+                                            )
+                                            # Release all locks and restart
+                                            for lock in locks.locks:
+                                                lock.release()
+                                            locks.locks.clear()
+                                            # Put failed dependency first
+                                            dynamic_deps.remove(dependency)
+                                            dynamic_deps.insert(0, dependency)
+                                            all_locked = False
+                                            break
+
+                                    if all_locked:
+                                        # All locks acquired successfully
+                                        break
+
+                        # Dependencies have been locked, we can start the job
+                        job.starttime = time.time()
+
+                        # Creates the main directory
+                        directory = job.path
+                        logger.debug("Making directories job %s...", directory)
+
+                        # Warn about directory cleanup for non-resumable tasks
+                        # (only once per task type)
+                        xpmtype = job.config.__xpmtype__
+                        if (
+                            directory.is_dir()
+                            and not job.resumable
+                            and not xpmtype.warned_clean_not_resumable
+                        ):
+                            xpmtype.warned_clean_not_resumable = True
+                            logger.warning(
+                                "In a future version, directory will be cleaned up for "
+                                "non-resumable tasks (%s). Use ResumableTask if you want "
+                                "to preserve the directory contents.",
+                                xpmtype.identifier,
+                            )
+
+                        if not directory.is_dir():
+                            directory.mkdir(parents=True, exist_ok=True)
+
+                        # Write metadata with submit and start time (after directory creation)
+                        job.write_metadata()
+
+                        # Sets up the notification URL
+                        if self.server is not None:
+                            job.add_notification_server(self.server)
+
+                    except Exception:
+                        logger.warning("Error while locking job", exc_info=True)
+                        return JobState.WAITING
+
+                    try:
+                        # Runs the job
+                        process = await job.aio_run()
+                    except Exception:
+                        logger.warning("Error while starting job", exc_info=True)
+                        return JobState.ERROR
+
+                # Wait for job to complete while holding locks
+                try:
+                    logger.debug("Waiting for job %s process to end", job)
+
+                    code = await process.aio_code()
+                    logger.debug("Got return code %s for %s", code, job)
+
+                    # Record exit code if available
+                    if code is not None:
+                        logger.info("Job %s ended with code %s", job, code)
+                        job.exit_code = code
+                    else:
+                        logger.info("Job %s ended, reading state from files", job)
+
+                    # Read state from .done/.failed files (contains detailed failure reason)
+                    state = JobState.from_path(job.path, job.name)
+
+                    # If state is a generic FAILED error, let the process determine
+                    # the state (it may detect launcher-specific failures like SLURM timeout)
+                    if (
+                        state is not None
+                        and isinstance(state, JobStateError)
+                        and state.failure_reason == JobFailureStatus.FAILED
+                        and code is not None
+                    ):
+                        process_state = process.get_job_state(code)
+                        if (
+                            isinstance(process_state, JobStateError)
+                            and process_state.failure_reason != JobFailureStatus.FAILED
+                        ):
+                            # Process detected a more specific failure reason
+                            state = process_state
+
+                    if state is None:
+                        if code is not None:
+                            # Fall back to process-specific state detection
+                            state = process.get_job_state(code)
+                        else:
+                            logger.error(
+                                "No .done or .failed file found for job %s", job
+                            )
+                            state = JobState.ERROR
+
+                except JobError:
+                    logger.warning("Error while running job")
+                    state = JobState.ERROR
+
+                except Exception:
+                    logger.warning(
+                        "Error while running job (in experimaestro)", exc_info=True
+                    )
+                    state = JobState.ERROR
+
+            # Locks are released here after job completes
+
+            # Check if we should restart a resumable task that timed out
+            from experimaestro.scheduler.jobs import JobStateError
+
+            if (
+                isinstance(state, JobStateError)
+                and state.failure_reason == JobFailureStatus.TIMEOUT
+                and job.resumable
+            ):
+                job.retry_count += 1
+                if job.retry_count <= job.max_retries:
+                    logger.info(
+                        "Resumable task %s timed out - restarting (attempt %d/%d)",
+                        job,
+                        job.retry_count,
+                        job.max_retries,
+                    )
+                    # Rotate log files to preserve previous run's logs
+                    job.rotate_logs()
+                    # Clear cached process so aio_run() will create a new one
+                    job._process = None
+                    # Delete PID file so the job will be resubmitted
+                    if job.pidpath.exists():
+                        job.pidpath.unlink()
+                    # Continue the loop to restart
+                    continue
+                else:
+                    logger.warning(
+                        "Resumable task %s exceeded max retries (%d), marking as failed",
+                        job,
+                        job.max_retries,
+                    )
+                    # Fall through to return the error state
+
+            # Job finished (success or non-recoverable error)
+            # Notify scheduler listeners of job state after job completes
             self.notify_job_state(job)
             return state
-
-        except Exception:
-            logger.warning("Error in scheduler job coordination", exc_info=True)
-            return JobState.ERROR