django-nativemojo 0.1.15__py3-none-any.whl → 0.1.16__py3-none-any.whl

mojo/apps/jobs/job_engine.py

@@ -0,0 +1,734 @@
+"""
+JobEngine - The runner daemon for executing jobs.
+
+Plan B engine: consumes jobs from Redis Lists (per-channel queues),
+tracks in-flight jobs in a ZSET with visibility timeout, and executes
+registered handlers.
+"""
+import sys
+import signal
+import socket
+import time
+import json
+import threading
+import random
+import traceback
+from datetime import datetime, timedelta
+from typing import Dict, List, Optional, Set, Tuple
+
+from django.db import close_old_connections
+
+from mojo.helpers.settings import settings
+from mojo.helpers import logit
+from .keys import JobKeys
+from .adapters import get_adapter
+from .models import Job, JobEvent
+import concurrent.futures
+import importlib
+from threading import Lock, Semaphore
+from typing import Callable
+
+from mojo.apps import metrics
+from mojo.helpers import dates
+
+logger = logit.get_logger("jobs", "jobs.log", debug=True)
+
+
+JOBS_ENGINE_CLAIM_BATCH = settings.get('JOBS_ENGINE_CLAIM_BATCH', 5)
+JOBS_CHANNELS = settings.get('JOBS_CHANNELS', ['default'])
+JOBS_ENGINE_MAX_WORKERS = settings.get('JOBS_ENGINE_MAX_WORKERS', 10)
+JOBS_ENGINE_CLAIM_BUFFER = settings.get('JOBS_ENGINE_CLAIM_BUFFER', 2)
+JOBS_RUNNER_HEARTBEAT_SEC = settings.get('JOBS_RUNNER_HEARTBEAT_SEC', 5)
+JOBS_VISIBILITY_TIMEOUT_MS = settings.get('JOBS_VISIBILITY_TIMEOUT_MS', 30000)
+
+
+def load_job_function(func_path: str) -> Callable:
+    """
+    Dynamically import a job function.
+    Example: 'mojo.apps.account.jobs.send_invite'
+    """
+    try:
+        module_path, func_name = func_path.rsplit('.', 1)
+        module = importlib.import_module(module_path)
+        return getattr(module, func_name)
+    except (ImportError, AttributeError, ValueError) as e:
+        raise ImportError(f"Cannot load job function '{func_path}': {e}")
+
+
+class JobEngine:
+    """
+    Job execution engine that runs as a daemon process.
+
+    Plan B: Consumes jobs from Redis List queues and executes handlers dynamically
+    with support for retries, cancellation, and parallel execution. Tracks in-flight
+    jobs in a ZSET to enable crash recovery via a reaper.
+    """
+
+    def __init__(self, channels: Optional[List[str]] = None,
+                 runner_id: Optional[str] = None,
+                 max_workers: Optional[int] = None):
+        """
+        Initialize the job engine.
+
+        Args:
+            channels: List of channels to consume from (default: from settings.JOBS_CHANNELS)
+            runner_id: Unique runner identifier (auto-generated if not provided)
+            max_workers: Maximum thread pool workers (default from settings)
+        """
+        self.channels = channels or JOBS_CHANNELS
+        self.runner_id = runner_id or self._generate_runner_id()
+        self.redis = get_adapter()
+        self.keys = JobKeys()
+
+        # Thread pool configuration
+        self.max_workers = max_workers or JOBS_ENGINE_MAX_WORKERS
+        self.executor = concurrent.futures.ThreadPoolExecutor(
+            max_workers=self.max_workers,
+            thread_name_prefix=f"JobWorker-{self.runner_id}"
+        )
+
+        # Track active jobs
+        self.active_jobs = {}
+        self.active_lock = Lock()
+
+        # Limit claimed jobs
+        claim_buffer = JOBS_ENGINE_CLAIM_BUFFER
+        self.max_claimed = self.max_workers * claim_buffer
+        self.claim_semaphore = Semaphore(self.max_claimed)
+
+        # Control flags
+        self.running = False
+        self.is_initialized = False
+        self.stop_event = threading.Event()
+
+        # Heartbeat thread
+        self.heartbeat_thread = None
+        self.heartbeat_interval = JOBS_RUNNER_HEARTBEAT_SEC
+
+        # Control channel listener
+        self.control_thread = None
+
+        # Stats
+        self.jobs_processed = 0
+        self.jobs_failed = 0
+        self.start_time = None
+
+        logger.info(f"JobEngine initialized: runner_id={self.runner_id}, "
+                  f"channels={self.channels}")
+
+    def _generate_runner_id(self) -> str:
+        """Generate a consistent runner ID based on hostname and channels."""
+        hostname = socket.gethostname()
+        # Clean hostname for use in ID (remove dots, make lowercase)
+        clean_hostname = hostname.lower().replace('.', '-').replace('_', '-')
+
+        # # Create a consistent suffix based on channels served
+        # channels_hash = hash(tuple(sorted(self.channels))) % 10000
+
+        return f"{clean_hostname}-engine"
+
+    def initialize(self):
+        if (self.is_initialized):
+            logger.warning("JobEngine already initialized")
+            return
+        self.is_initialized = True
+
+        logger.info(f"Initializing JobEngine {self.runner_id}")
+        self.running = True
+        self.start_time = dates.utcnow()
+        self.stop_event.clear()
+
+        # Start heartbeat thread
+        self._start_heartbeat()
+
+        # Start control listener thread
+        self._start_control_listener()
+
+        # Register signal handlers
+        self._setup_signal_handlers()
+
+    def start(self):
+        """
+        Start the job engine.
+
+        Sets up consumer groups, starts heartbeat, and begins processing.
+        """
+        if self.running:
+            logger.warning("JobEngine already running")
+            return
+
+        self.initialize()
+
+        # Main processing loop
+        try:
+            self._main_loop()
+        except KeyboardInterrupt:
+            logger.info("JobEngine interrupted by user")
+        except Exception as e:
+            logger.error(f"JobEngine crashed: {e}")
+            raise
+        finally:
+            self.stop()
+
+    def stop(self, timeout: float = 30.0):
+        """
+        Stop the job engine gracefully.
+
+        Args:
+            timeout: Maximum time to wait for clean shutdown
+        """
+        if self.running:
+            logger.info(f"Stopping JobEngine {self.runner_id}...")
+            self.running = False
+            self.stop_event.set()
+            # Wait for active jobs
+            with self.active_lock:
+                active = list(self.active_jobs.values())
+            if active:
+                logger.info(f"Waiting for {len(active)} active jobs...")
+                futures = [j['future'] for j in active]
+                concurrent.futures.wait(futures, timeout=timeout/2)
+            # Shutdown executor
+            self.executor.shutdown(wait=True)
+
+        # Stop heartbeat
+        if self.heartbeat_thread and self.heartbeat_thread.is_alive():
+            self.heartbeat_thread.join(timeout=5.0)
+
+        # Stop control listener
+        if self.control_thread and self.control_thread.is_alive():
+            self.control_thread.join(timeout=5.0)
+
+        # Clean up Redis keys
+        try:
+            self.redis.delete(self.keys.runner_hb(self.runner_id))
+        except Exception as e:
+            logger.warning(f"Failed to clean up runner keys: {e}")
+
+        logger.info(f"JobEngine {self.runner_id} stopped. "
+                  f"Processed: {self.jobs_processed}, Failed: {self.jobs_failed}")
+
+    def _cleanup_consumer_groups(self):
+        """
+        Clean up consumer group registrations on shutdown.
+        This prevents accumulation of dead consumers.
+        """
+        logger.info(f"Cleaning up consumer registrations for {self.runner_id}")
+
+        for channel in self.channels:
+            try:
+                stream_key = self.keys.stream(channel)
+                group_key = self.keys.group_workers(channel)
+                broadcast_stream = self.keys.stream_broadcast(channel)
+                runner_group = self.keys.group_runner(channel, self.runner_id)
+
+                client = self.redis.get_client()
+
+                # For main stream: reclaim and ACK any pending jobs before deletion
+                try:
+                    pending_info = client.execute_command(
+                        'XPENDING', stream_key, group_key, '-', '+', '100', self.runner_id
+                    )
+
+                    if pending_info:
+                        message_ids = [msg[0] for msg in pending_info]
+                        if message_ids:
+                            # Reclaim and immediately ACK to clear them
+                            try:
+                                claimed = client.execute_command(
+                                    'XCLAIM', stream_key, group_key, self.runner_id,
+                                    '0', *message_ids
+                                )
+                                if claimed:
+                                    client.execute_command('XACK', stream_key, group_key, *message_ids)
+                                    logger.info(f"Cleared {len(message_ids)} pending jobs during cleanup for {channel}")
+                            except Exception as e:
+                                logger.warning(f"Failed to clear pending jobs during cleanup: {e}")
+
+                except Exception as e:
+                    logger.debug(f"No pending jobs to clean for {channel}: {e}")
+
+                # Delete consumer from main group
+                try:
+                    client.execute_command('XGROUP', 'DELCONSUMER', stream_key, group_key, self.runner_id)
+                    logger.debug(f"Removed consumer {self.runner_id} from group {group_key}")
+                except Exception as e:
+                    logger.debug(f"Consumer {self.runner_id} was not in group {group_key}: {e}")
+
+                # Delete consumer from broadcast group
+                try:
+                    client.execute_command('XGROUP', 'DELCONSUMER', broadcast_stream, runner_group, self.runner_id)
+                    logger.debug(f"Removed consumer {self.runner_id} from broadcast group {runner_group}")
+                except Exception as e:
+                    logger.debug(f"Consumer {self.runner_id} was not in broadcast group {runner_group}: {e}")
+
+            except Exception as e:
+                logger.warning(f"Failed to cleanup consumer groups for {channel}: {e}")
+
+    def _setup_consumer_groups(self):
+        """No-op in Plan B (List + ZSET)."""
+        logger.info("Plan B mode: no consumer groups to set up.")
+
+    def _setup_signal_handlers(self):
+        """Register signal handlers for graceful shutdown."""
+        def handle_signal(signum, frame):
+            logger.info(f"Received signal {signum}, initiating graceful shutdown")
+            self.stop()
+            sys.exit(0)
+
+        signal.signal(signal.SIGTERM, handle_signal)
+        signal.signal(signal.SIGINT, handle_signal)
+
+    def _start_heartbeat(self):
+        """Start the heartbeat and reaper threads."""
+        self.heartbeat_thread = threading.Thread(
+            target=self._heartbeat_loop,
+            name=f"Heartbeat-{self.runner_id}",
+            daemon=True
+        )
+        self.heartbeat_thread.start()
+        # Reaper thread for visibility timeout
+        self.reaper_thread = threading.Thread(
+            target=self._reaper_loop,
+            name=f"Reaper-{self.runner_id}",
+            daemon=True
+        )
+        self.reaper_thread.start()
+
+    def _heartbeat_loop(self):
+        """Heartbeat thread main loop."""
+        hb_key = self.keys.runner_hb(self.runner_id)
+
+        while self.running and not self.stop_event.is_set():
+            try:
+                # Update heartbeat with TTL
+                self.redis.set(hb_key, json.dumps({
+                    'runner_id': self.runner_id,
+                    'hostname': socket.gethostname(),
+                    'channels': self.channels,
+                    'jobs_processed': self.jobs_processed,
+                    'jobs_failed': self.jobs_failed,
+                    'started': self.start_time.isoformat(),
+                    'last_heartbeat': dates.utcnow().isoformat()
+                }), ex=self.heartbeat_interval * 3)  # TTL = 3x interval
+
+                # Touch visibility timeout for active jobs to prevent premature reaping
+                try:
+                    now_ms = int(time.time() * 1000)
+                    # Snapshot active jobs to minimize lock hold time
+                    with self.active_lock:
+                        active_snapshot = [(jid, meta.get('channel')) for jid, meta in self.active_jobs.items()]
+                    for jid, ch in active_snapshot:
+                        if not ch:
+                            continue
+                        # Update in-flight ZSET score to extend visibility timeout
+                        self.redis.zadd(self.keys.processing(ch), {jid: now_ms})
+                except Exception as te:
+                    logger.debug(f"Heartbeat touch failed: {te}")
+
+            except Exception as e:
+                logger.warning(f"Heartbeat update failed: {e}")
+
+            # Sleep with periodic wake for stop check
+            for _ in range(self.heartbeat_interval):
+                if self.stop_event.is_set():
+                    break
+                time.sleep(1)
+
+    def _start_control_listener(self):
+        """Start the control channel listener thread."""
+        self.control_thread = threading.Thread(
+            target=self._control_loop,
+            name=f"Control-{self.runner_id}",
+            daemon=True
+        )
+        self.control_thread.start()
+
+    def _control_loop(self):
+        """Control channel listener loop."""
+        control_key = self.keys.runner_ctl(self.runner_id)
+        broadcast_key = "mojo:jobs:runners:broadcast"
+        pubsub = self.redis.pubsub()
+        # Listen to runner-specific control and global broadcast control
+        pubsub.subscribe(control_key, broadcast_key)
+
+        try:
+            while self.running and not self.stop_event.is_set():
+                message = pubsub.get_message(timeout=5.0)
+                if message and message.get('type') == 'message':
+                    self._handle_control_message(message.get('data'), message.get('channel'))
+        finally:
+            pubsub.close()
+
+    def _handle_control_message(self, data: bytes, channel: Optional[str] = None):
+        """Handle a control channel message or broadcast command."""
+        try:
+            message = json.loads(data.decode('utf-8'))
+            command = message.get('command')
+
+            if command == 'ping':
+                # Respond with pong (direct control)
+                response_key = message.get('response_key')
+                if response_key:
+                    self.redis.set(response_key, 'pong', ex=5)
+                logger.info("Responded to ping from control channel")
+
+            elif command == 'status':
+                # Broadcast status reply
+                reply_channel = message.get('reply_channel')
+                if reply_channel:
+                    reply = {
+                        'runner_id': self.runner_id,
+                        'channels': self.channels,
+                        'jobs_processed': self.jobs_processed,
+                        'jobs_failed': self.jobs_failed,
+                        'started': self.start_time.isoformat() if self.start_time else None,
+                        'timestamp': dates.utcnow().isoformat(),
+                    }
+                    try:
+                        self.redis.publish(reply_channel, json.dumps(reply))
+                    except Exception as e:
+                        logger.warning(f"Failed to publish status reply: {e}")
+
+            elif command == 'shutdown':
+                logger.info("Received shutdown command from control channel/broadcast")
+                self.stop()
+
+            else:
+                logger.warning(f"Unknown control command: {command}")
+
+        except Exception as e:
+            logger.error(f"Failed to handle control message: {e}")
+
+    def _main_loop(self):
+        """Main processing loop - claims jobs from List queues based on capacity."""
+        logger.info(f"JobEngine {self.runner_id} entering main loop (Plan B)")
+
+        while self.running and not self.stop_event.is_set():
+            try:
+                # Check available capacity
+                with self.active_lock:
+                    active_count = len(self.active_jobs)
+
+                if active_count >= self.max_claimed:
+                    time.sleep(0.1)
+                    continue
+
+                # Compose BRPOP order (priority first)
+                channels_ordered = list(self.channels)
+                if 'priority' in channels_ordered:
+                    channels_ordered = ['priority'] + [c for c in channels_ordered if c != 'priority']
+                queue_keys = [self.keys.queue(ch) for ch in channels_ordered]
+
+                # Claim one job at a time to avoid over-claiming
+                popped = self.redis.brpop(queue_keys, timeout=1)
+                if not popped:
+                    continue
+
+                queue_key, job_id = popped
+                # Determine channel from key
+                channel = queue_key.split(':')[-1]
+
+                # Track in-flight (visibility)
+                try:
+                    self.redis.zadd(self.keys.processing(channel), {job_id: int(time.time() * 1000)})
+                except Exception as e:
+                    logger.warning(f"Failed to add job {job_id} to processing ZSET: {e}")
+
+                # Submit to thread pool
+                future = self.executor.submit(
+                    self.execute_job,
+                    channel, job_id
+                )
+
+                with self.active_lock:
+                    self.active_jobs[job_id] = {
+                        'future': future,
+                        'started': dates.utcnow(),
+                        'channel': channel
+                    }
+
+                future.add_done_callback(lambda f, jid=job_id: self._job_completed(jid))
+
+            except Exception as e:
+                logger.error(f"Error in main loop: {e}")
+                time.sleep(0.5)
+
+    def claim_jobs_by_channel(self, channel: str, count: int) -> List[Tuple[str, str, str]]:
+        """Plan B: not used. Kept for compatibility."""
+        return []
+
+    def claim_jobs(self, count: int) -> List[Tuple[str, str, str]]:
+        """
+        Claim up to 'count' jobs from Redis streams.
+
+        Args:
+            count: Maximum number of jobs to claim
+
+        Returns:
+            List of (stream_key, msg_id, job_id) tuples
+        """
+        claimed = []
+        # Prioritize 'priority' channel first if present
+        channels_ordered = list(self.channels)
+        if 'priority' in channels_ordered:
+            channels_ordered = ['priority'] + [c for c in channels_ordered if c != 'priority']
+        for channel in channels_ordered:
+            if len(claimed) >= count:
+                break
+            channel_messages = self.claim_jobs_by_channel(channel, count - len(claimed))
+            claimed.extend(channel_messages)
+        return claimed
+
+    def _ack_message(self, stream_key: str, msg_id: str):
+        """Plan B: not used. Kept for compatibility."""
+        return
+
+    def execute_job(self, channel: str, job_id: str):
+        """Execute job and handle all state updates (Plan B)."""
+        job = None
+        try:
+            # Load job from database
+            close_old_connections()
+            job = Job.objects.select_for_update().get(id=job_id)
+        except Exception as e:
+            logit.error(f"Failed to load job {job_id}: {e}")
+            # Remove from processing to avoid leak
+            try:
+                self.redis.zrem(self.keys.processing(channel), job_id)
+            except Exception:
+                pass
+            return
+
+        try:
+            # Check if already processed or canceled
+            if job.status in ('completed', 'canceled'):
+                # Already finished; remove from processing if present
+                try:
+                    self.redis.zrem(self.keys.processing(channel), job_id)
+                except Exception:
+                    pass
+                return
+
+            # Check expiration
+            if job.is_expired:
+                job.status = 'expired'
+                job.finished_at = dates.utcnow()
+                job.save(update_fields=['status', 'finished_at'])
+
+                # Event: expired
+                try:
+                    JobEvent.objects.create(
+                        job=job,
+                        channel=job.channel,
+                        event='expired',
+                        runner_id=self.runner_id,
+                        attempt=job.attempt,
+                        details={'reason': 'job_expired_before_execution'}
+                    )
+                except Exception:
+                    pass
+
+                # Remove from processing after DB update
+                try:
+                    self.redis.zrem(self.keys.processing(channel), job_id)
+                except Exception:
+                    pass
+                metrics.record("jobs.expired")
+                return
+
+            # Mark as running
+            job.status = 'running'
+            job.started_at = dates.utcnow()
+            job.runner_id = self.runner_id
+            job.attempt += 1
+            job.save(update_fields=['status', 'started_at', 'runner_id', 'attempt'])
+
+            # Event: running
+            try:
+                JobEvent.objects.create(
+                    job=job,
+                    channel=job.channel,
+                    event='running',
+                    runner_id=self.runner_id,
+                    attempt=job.attempt,
+                    details={'queue': self.keys.queue(channel)}
+                )
+            except Exception:
+                pass
+
+            # Load and execute function
+            func = load_job_function(job.func)
+            func(job)
+
+            # Mark complete
+            job.status = 'completed'
+            job.finished_at = dates.utcnow()
+            job.save(update_fields=['status', 'finished_at', 'metadata'])
+            logger.info(f"Job {job.id} completed")
+            # Event: completed
+            try:
+                JobEvent.objects.create(
+                    job=job,
+                    channel=job.channel,
+                    event='completed',
+                    runner_id=self.runner_id,
+                    attempt=job.attempt,
+                    details={}
+                )
+            except Exception:
+                pass
+
+            # Remove from processing after DB update
+            try:
+                self.redis.zrem(self.keys.processing(channel), job_id)
+            except Exception:
+                pass
+
+            # Metrics
+            metrics.record("jobs.completed", count=1)
+            metrics.record(f"jobs.channel.{job.channel}.completed", count=1)
+            metrics.record("jobs.duration_ms", count=job.duration_ms)
+
+        except Exception as e:
+            try:
+                if job:
+                    job.add_log(f"Failed to complete job: {e}", kind="error")
+            except Exception:
+                pass
+            self._handle_job_failure(job_id, channel, e)
+
+    def _handle_job_failure(self, job_id: str, channel: str, error: Exception):
+        """Handle job failure with retries (Plan B)."""
+        try:
+            job = Job.objects.select_for_update().get(id=job_id)
+
+            # Record error
+            job.last_error = str(error)
+            job.stack_trace = traceback.format_exc()
+
+            # Check retry eligibility
+            if job.attempt < job.max_retries:
+                # Calculate backoff with jitter
+                backoff = min(
+                    job.backoff_base ** job.attempt,
+                    job.backoff_max_sec
+                )
+                jitter = backoff * (0.8 + random.random() * 0.4)
+
+                # Schedule retry
+                job.run_at = dates.utcnow() + timedelta(seconds=jitter)
+                job.status = 'pending'
+                job.save(update_fields=[
+                    'status', 'run_at', 'last_error', 'stack_trace'
+                ])
+
+                # Event: retry scheduled
+                try:
+                    JobEvent.objects.create(
+                        job=job,
+                        channel=job.channel,
+                        event='retry',
+                        runner_id=self.runner_id,
+                        attempt=job.attempt,
+                        details={'reason': 'failure', 'next_run_at': job.run_at.isoformat()}
+                    )
+                except Exception:
+                    pass
+
+                # Add to scheduled ZSET (route by broadcast)
+                score = job.run_at.timestamp() * 1000
+                target_zset = self.keys.sched_broadcast(job.channel) if job.broadcast else self.keys.sched(job.channel)
+                self.redis.zadd(target_zset, {job_id: score})
+
+                metrics.record("jobs.retried")
+            else:
+                # Max retries exceeded
+                job.status = 'failed'
+                job.finished_at = dates.utcnow()
+                job.save(update_fields=[
+                    'status', 'finished_at', 'last_error', 'stack_trace'
+                ])
+
+                # Event: failed
+                try:
+                    JobEvent.objects.create(
+                        job=job,
+                        channel=job.channel,
+                        event='failed',
+                        runner_id=self.runner_id,
+                        attempt=job.attempt,
+                        details={'error': job.last_error}
+                    )
+                except Exception:
+                    pass
+
+                metrics.record("jobs.failed")
+                metrics.record(f"jobs.channel.{job.channel}.failed")
+
+            # Always remove from processing to prevent leaks
+            try:
+                self.redis.zrem(self.keys.processing(channel), job_id)
+            except Exception:
+                pass
+
+        except Exception as e:
+            logit.error(f"Failed to handle job failure: {e}")
+
+    def _job_completed(self, job_id: str):
+        """Callback when job future completes."""
+        with self.active_lock:
+            self.active_jobs.pop(job_id, None)
+        self.jobs_processed += 1
+
+    def _reaper_loop(self):
+        """Requeue stale in-flight jobs based on visibility timeout (Plan B)."""
+        while self.running and not self.stop_event.is_set():
+            try:
+                now_ms = int(time.time() * 1000)
+                cutoff = now_ms - JOBS_VISIBILITY_TIMEOUT_MS
+                for ch in self.channels:
+                    # Acquire short-lived lock to avoid duplicate requeues across engines
+                    acquired = False
+                    try:
+                        acquired = self.redis.set(self.keys.reaper_lock(ch), self.runner_id, nx=True, px=2000)
+                    except Exception as le:
+                        logger.debug(f"Reaper lock error for {ch}: {le}")
+                        acquired = False
+                    if not acquired:
+                        # Another engine is handling this channel right now
+                        continue
+                    # Fetch stale entries: claimed earlier than cutoff
+                    try:
+                        stale_ids = self.redis.zrangebyscore(self.keys.processing(ch), float("-inf"), cutoff, limit=100)
+                    except Exception as e:
+                        logger.debug(f"Reaper fetch failed for {ch}: {e}")
+                        stale_ids = []
+                    for jid in stale_ids:
+                        try:
+                            # Remove from processing and requeue
+                            self.redis.zrem(self.keys.processing(ch), jid)
+                            self.redis.rpush(self.keys.queue(ch), jid)
+                            # Add event trail (best effort)
+                            try:
+                                job = Job.objects.get(id=jid)
+                                JobEvent.objects.create(
+                                    job=job,
+                                    channel=ch,
+                                    event='retry',
+                                    runner_id=self.runner_id,
+                                    attempt=job.attempt,
+                                    details={'reason': 'reaper_timeout'}
+                                )
+                            except Exception:
+                                pass
+                            logger.info(f"Reaper requeued stale job {jid} on {ch}")
+                        except Exception as e:
+                            logger.warning(f"Reaper failed to requeue {jid} on {ch}: {e}")
+            except Exception as e:
+                logger.warning(f"Reaper loop error: {e}")
+            # Sleep a bit before next pass
+            for _ in range(5):
+                if self.stop_event.is_set():
+                    break
+                time.sleep(1)