Work: Background Tasks

Module: sillo.work.background Source files:

• /Users/admin/sillo.build/core/sillo/work/background/tasks.py (235 lines)
• /Users/admin/sillo.build/core/sillo/work/background/supervisor.py (137 lines)
• /Users/admin/sillo.build/core/sillo/work/__init__.py (73 lines)
• /Users/admin/sillo.build/core/sillo/work/types.py (300 lines)

Version: 2026-08-11 Audience: Core maintainers, framework architects Purpose: Deep documentation of BackgroundTask, Supervisor, restart policies, circuit breaker, health/stats, and setup_work wiring

---

1. Overview

The background task subsystem provides fire-and-forget async execution with lifecycle tracking, supervision, and graceful shutdown. It sits above the Task class and provides a simpler API for common patterns.

graph TD
    A["BackgroundTask.run(func, ...)"] -->|creates| B[Task instance]
    B -->|wrapped in| C["asyncio.ensure_future()"]
    C -->|tracked in| D["_instances class set"]
    D -->|drained by| E["BackgroundTask.drain()"]

    F[Supervisor] -->|monitors| A
    F -->|restarts on| G{failure?}
    G -->|Yes| H{"_should_restart()?"}
    H -->|Yes| I["sleep(delay)"] --> A
    H -->|No| J[Supervisor stops]

---

2. BackgroundTask

File: /Users/admin/sillo.build/core/sillo/work/background/tasks.py (235 lines)

2.1 Class-Level Tracking

class BackgroundTask:
    _instances: ClassVar[set[BackgroundTask]] = set()
    _lock = asyncio.Lock()

Every BackgroundTask instance is automatically added to _instances on construction and (implicitly) removed when garbage collected. This enables:

• drain(): wait for all tracked tasks before shutdown
• count(): monitoring dashboard metrics

2.2 Constructor

def __init__(
    self,
    func: Callable[..., Awaitable[Any]],
    *args: Any,
    name: str | None = None,
    on_done: Callable[[TaskResult], Awaitable[None]] | None = None,
    on_success: Callable[[TaskResult], Awaitable[None]] | None = None,
    on_failure: Callable[[TaskResult], Awaitable[None]] | None = None,
    timeout: float | None = None,
    metadata: dict[str, Any] | None = None,
    **kwargs: Any,
) -> None:

Key behavior:

1. Creates a Task instance internally
2. Attaches on_done as both success and failure callback
3. Attaches on_success and on_failure individually
4. Launches immediately via asyncio.ensure_future(self._task_obj.run())
5. Adds self to BackgroundTask._instances

self._task_obj = Task(func, *args, name=name or func.__name__, metadata=metadata, timeout=timeout, **kwargs)
if on_done:
    self._task_obj.on_success(on_done).on_failure(on_done)
if on_success:
    self._task_obj.on_success(on_success)
if on_failure:
    self._task_obj.on_failure(on_failure)
self._asyncio_task = asyncio.ensure_future(self._task_obj.run())
BackgroundTask._instances.add(self)

2.3 Properties

Property	Type	Description
done	bool	True if completed, failed, or cancelled
running	bool	True if currently executing
result	TaskResult | None	Result if completed, else None
id	str	UUID4 task identifier
name	str	Human-readable task name
elapsed	float	Seconds since launch

2.4 wait()

async def wait(self, timeout: float | None = None) -> Any:
    return await self._task_obj.wait(timeout=timeout)

Blocks until the task completes. Raises the original exception if the task failed.

2.5 cancel()

def cancel(self) -> bool:
    if self._asyncio_task and not self._asyncio_task.done():
        return self._asyncio_task.cancel()
    return False

2.6 run(): Class Method Factory

@classmethod
def run(cls, func, *args, **kwargs) -> BackgroundTask:
    try:
        asyncio.get_running_loop()
    except RuntimeError:
        raise RuntimeError("BackgroundTask.run() requires an async context")
    return cls(func, *args, **kwargs)

Must be called from within an async context (running event loop).

2.7 run_sync(): Auto-Wrap Sync Functions

@classmethod
def run_sync(cls, func, *args, **kwargs) -> BackgroundTask:
    if not inspect.iscoroutinefunction(func):
        async def _wrapper(*a, **kw):
            return func(*a, **kw)
        return cls(_wrapper, *args, **kwargs)
    return cls(func, *args, **kwargs)

Automatically wraps synchronous functions in an async wrapper.

2.8 drain(): Graceful Shutdown

@classmethod
async def drain(cls, timeout=10.0, cancel_remaining=True) -> dict:
    instances = list(cls._instances)
    if not instances:
        return {"total": 0, "completed": 0, "cancelled": 0}

    tasks = [i._asyncio_task for i in instances]
    done, pending = await asyncio.wait(tasks, timeout=timeout)

    cancelled = 0
    if cancel_remaining and pending:
        for t in pending:
            t.cancel()
        cancelled = len(pending)

    return {
        "total": len(instances),
        "completed": len(done) - cancelled,
        "cancelled": cancelled,
    }

sequenceDiagram
    participant App as Application
    participant BT as BackgroundTask
    participant T1 as Task 1
    participant T2 as Task 2
    participant T3 as Task 3

    App->>BT: drain(timeout=10)
    BT->>T1: wait
    BT->>T2: wait
    BT->>T3: wait
    Note over BT: asyncio.wait(timeout=10)
    T1-->>BT: done
    T2-->>BT: done
    Note over T3: Still running after 10s
    BT->>T3: cancel()
    BT-->>App: {"total": 3, "completed": 2, "cancelled": 1}

2.9 count(): Status Summary

@classmethod
def count(cls) -> dict[str, int]:
    total = len(cls._instances)
    running = sum(1 for t in cls._instances if t.running)
    done = sum(1 for t in cls._instances if t.done)
    return {
        "total": total,
        "running": running,
        "done": done,
        "pending": total - running - done,
    }

2.10 to_dict()

def to_dict(self) -> dict[str, Any]:
    return {
        "id": self.id,
        "name": self.name,
        "done": self.done,
        "running": self.running,
        "elapsed": self.elapsed,
        "status": self._task_obj.status.value,
        "result": self.result.to_dict() if self.result else None,
    }

---

3. Supervisor

File: /Users/admin/sillo.build/core/sillo/work/background/supervisor.py (137 lines)

3.1 RestartPolicy Enum

class RestartPolicy(Enum):
    NEVER = "never"
    ALWAYS = "always"
    ON_FAILURE = "on_failure"
    EXPONENTIAL_BACKOFF = "exponential_backoff"

Policy	Behavior
NEVER	Task runs once; no restarts
ALWAYS	Restart on success or failure (up to max_restarts)
ON_FAILURE	Restart only on failure (up to max_restarts)
EXPONENTIAL_BACKOFF	Restart on failure with exponential delay

3.2 Supervisor Class

class Supervisor:
    def __init__(
        self,
        func: Callable[..., Awaitable[Any]],
        policy: RestartPolicy = RestartPolicy.ON_FAILURE,
        *,
        max_restarts: int = 3,
        base_delay: float = 1.0,
        max_delay: float = 60.0,
        name: str | None = None,
    ):
        self.func = func
        self.policy = policy
        self.max_restarts = max_restarts
        self.base_delay = base_delay
        self.max_delay = max_delay
        self.name = name or func.__name__
        self._restarts = 0
        self._current_task: BackgroundTask | None = None
        self._running = False
        self._stopped = asyncio.Event()

3.3 start(): Monitoring Loop

async def start(self, *args, **kwargs) -> None:
    self._running = True
    self._restarts = 0
    self._stopped.clear()

    while self._running:
        self._current_task = BackgroundTask.run(self.func, *args, name=self.name, **kwargs)
        try:
            await self._current_task.wait()
        except asyncio.CancelledError:
            break
        except Exception as exc:
            logger.warning("Supervised task %s failed: %s", self.name, exc)
            if not self._should_restart():
                logger.error("Supervised task %s exhausted restarts (%d)", self.name, self.max_restarts)
                break
            delay = min(self.base_delay * (2 ** self._restarts), self.max_delay)
            logger.info("Restarting %s in %.1fs (attempt %d)", self.name, delay, self._restarts + 1)
            await asyncio.sleep(delay)
            self._restarts += 1
        else:
            if self.policy == RestartPolicy.NEVER or self.policy == RestartPolicy.ON_FAILURE:
                break

    self._stopped.set()

flowchart TD
    A["start()"] --> B["while running"]
    B --> C["BackgroundTask.run(func)"]
    C --> D["await task.wait()"]
    D --> E{Result?}
    E -->|CancelledError| F[Exit loop]
    E -->|Exception| G{"_should_restart()?"}
    G -->|No| H[Log exhausted, exit]
    G -->|Yes| I["delay = base * 2^restarts"]
    I --> J["sleep(delay)"]
    J --> K["restarts += 1"]
    K --> B
    E -->|Success| L{Policy?}
    L -->|NEVER or ON_FAILURE| F
    L -->|ALWAYS| B
    F --> M["_stopped.set()"]

3.4 _should_restart()

def _should_restart(self) -> bool:
    if self.policy == RestartPolicy.NEVER:
        return False
    if self.policy == RestartPolicy.ALWAYS:
        return self.max_restarts == 0 or self._restarts < self.max_restarts
    if self.policy in (RestartPolicy.ON_FAILURE, RestartPolicy.EXPONENTIAL_BACKOFF):
        return self.max_restarts == 0 or self._restarts < self.max_restarts
    return False

Note: max_restarts=0 means unlimited restarts.

3.5 stop() and wait()

def stop(self) -> None:
    self._running = False
    if self._current_task:
        self._current_task.cancel()

async def wait(self, timeout: float | None = None) -> None:
    await asyncio.wait_for(self._stopped.wait(), timeout=timeout)

3.6 Exponential Backoff Calculation

delay = min(self.base_delay * (2 ** self._restarts), self.max_delay)

Restart #	Delay (base=1.0)	Delay (base=2.0)
0	1.0s	2.0s
1	2.0s	4.0s
2	4.0s	8.0s
3	8.0s	16.0s
4	16.0s	32.0s
5	32.0s	60.0s (capped)

---

4. Circuit Breaker

File: /Users/admin/sillo.build/core/sillo/work/types.py

4.1 CircuitState Enum

class CircuitState(enum.Enum):
    CLOSED = "closed"      # Normal operation
    OPEN = "open"          # Failures exceeded threshold
    HALF_OPEN = "half_open"  # Testing recovery

stateDiagram-v2
    [*] --> CLOSED
    CLOSED --> OPEN : failures > threshold
    OPEN --> HALF_OPEN : timeout elapsed
    HALF_OPEN --> CLOSED : success
    HALF_OPEN --> OPEN : failure

The circuit breaker pattern prevents cascading failures:

• CLOSED: Normal operation; requests flow through
• OPEN. Failures exceeded threshold; requests are shed immediately
• HALF_OPEN: After a timeout, one test request is allowed through

4.2 CircuitBreakerOpen Exception

class CircuitBreakerOpen(WorkError):
    """Worker circuit is open — requests are being shed."""

---

5. Health and Stats

5.1 QueueHealth

class QueueHealth(enum.Enum):
    HEALTHY = "healthy"
    DEGRADED = "degraded"  # High backlog
    STALLED = "stalled"    # No consumers

5.2 QueueStats

@dataclasses.dataclass
class QueueStats:
    name: str
    size: int = 0
    completed: int = 0
    failed: int = 0
    oldest_age_ms: int = 0
    status: QueueHealth = QueueHealth.HEALTHY

5.3 WorkerStats

@dataclasses.dataclass
class WorkerStats:
    processed: int = 0
    failed: int = 0
    active: int = 0
    workers: int = 0
    circuit: CircuitState = CircuitState.CLOSED
    uptime_seconds: float = 0.0

5.4 SchedulerStats

@dataclasses.dataclass
class SchedulerStats:
    jobs_total: int = 0
    jobs_active: int = 0
    jobs_paused: int = 0
    runs_total: int = 0
    errors_total: int = 0

5.5 TaskResult

@dataclasses.dataclass(frozen=True, slots=True)
class TaskResult:
    task_id: str
    name: str
    status: TaskStatus
    result: Any = None
    error: str | None = None
    attempt: int = 0
    max_attempts: int = 0
    priority: TaskPriority = TaskPriority.NORMAL
    queue_name: str = "default"
    created_at: float = ...
    started_at: float = 0.0
    completed_at: float = 0.0
    metadata: dict[str, Any] = ...
    worker_id: str = ""

---

6. setup_work() Wiring

File: /Users/admin/sillo.build/core/sillo/work/__init__.py

def setup_work(app, *, queue_backend=None, queue_name="default") -> dict:

6.1 Wiring Steps

1. Create connection: SyncConnection() or RedisConnection() based on queue_backend
2. Store in app.state: app.state["queue_connection"], app.state["default_queue"]
3. Register DI providers: scheduler, queue_connection, events, default_queue
4. Hook lifecycle: Scheduler start/stop via app.on_startup / app.on_shutdown
5. Import commands: sillo.work.commands at module level for CLI registration

6.2 DI Providers

File: /Users/admin/sillo.build/core/sillo/work/dependency.py

scheduler = _make_provider("scheduler")
queue_connection = _make_provider("queue_connection")
events = _make_provider("events")
default_queue = _make_provider("default_queue")

Each provider is an async function that pulls from app.state via the DI system.

6.3 Usage

from sillo.work import setup_work

app = SilloApp()
work = setup_work(app, queue_backend="redis")

# Or with defaults (in-memory)
work = setup_work(app)

---

7. Usage Patterns

7.1 Fire-and-Forget

bt = BackgroundTask.run(send_email, user.email)
# Don't wait — just fire and forget

7.2 With Completion Callback

async def notify(result: TaskResult):
    await send_notification(f"Task {result.name} completed: {result.status.value}")

bt = BackgroundTask.run(process_file, path, on_done=notify)

7.3 Supervised Task

supervisor = Supervisor(
    websocket_listener,
    RestartPolicy.EXPONENTIAL_BACKOFF,
    max_restarts=5,
    base_delay=1.0,
    max_delay=60.0,
)
await supervisor.start(url="wss://stream.example.com")

7.4 Graceful Shutdown

@app.on_shutdown
async def shutdown():
    result = await BackgroundTask.drain(timeout=30.0, cancel_remaining=True)
    logger.info("Drained: %s", result)

---

8. Design Decisions

D-1: Class-Level Instance Tracking

Using a class-level set enables drain() and count() without requiring a global registry. The trade-off is that instances are never explicitly removed from the set. They are garbage collected when no references remain.

D-2: Immediate Launch on Construction

BackgroundTask launches immediately on construction (via asyncio.ensure_future). This is intentional for fire-and-forget patterns but means the task cannot be configured after construction.

D-3: Supervisor as a Separate Class

The Supervisor is decoupled from BackgroundTask so it can monitor any async callable, not just tasks created through the background task API.

D-4: Exponential Backoff with Cap

The min(base * 2^n, max_delay) formula prevents unbounded delay growth while still providing meaningful backoff for transient failures.

---

9. Source Traceability

Component	File	Lines
BackgroundTask	core/sillo/work/background/tasks.py	27-235
RestartPolicy enum	core/sillo/work/background/supervisor.py	24-30
Supervisor	core/sillo/work/background/supervisor.py	33-137
CircuitState enum	core/sillo/work/types.py	75-80
QueueHealth enum	core/sillo/work/types.py	83-88
QueueStats	core/sillo/work/types.py	236-256
WorkerStats	core/sillo/work/types.py	259-279
SchedulerStats	core/sillo/work/types.py	282-300
TaskResult	core/sillo/work/types.py	138-233
setup_work()	core/sillo/work/__init__.py	1-73
DI providers	core/sillo/work/dependency.py	1-49

---

10. BackgroundTask Internal Mechanics

10.1 Instance Lifecycle

sequenceDiagram
    participant U as User Code
    participant BT as BackgroundTask
    participant T as Task
    participant AF as asyncio.Future

    U->>BT: BackgroundTask(func, *args)
    BT->>T: Task(func, *args)
    BT->>BT: Attach callbacks
    BT->>AF: asyncio.ensure_future(task.run())
    BT->>BT: _instances.add(self)
    Note over AF: Running in background
    AF-->>BT: done (via _done Event)
    BT->>T: task.result available

10.2 The _instances Set

_instances: ClassVar[set[BackgroundTask]] = set()

Key properties:

• Class-level, shared across all BackgroundTask instances
• Instances are added on construction
• Instances are not explicitly removed: they are garbage collected when no external references remain
• drain() snapshots the set at call time: instances = list(cls._instances)
• count() iterates the set to compute status breakdowns

Memory consideration: In long-running applications with many fire-and-forget tasks, the _instances set may hold references to completed tasks. If this becomes a concern, periodically call count() or drain() which implicitly triggers cleanup through Python’s garbage collector.

10.3 Callback Attachment

if on_done:
    self._task_obj.on_success(on_done).on_failure(on_done)
if on_success:
    self._task_obj.on_success(on_success)
if on_failure:
    self._task_obj.on_failure(on_failure)

The on_done callback is registered for both success and failure. The on_success and on_failure callbacks are registered individually. All callbacks receive a TaskResult instance.

10.4 Sync Function Wrapping

@classmethod
def run_sync(cls, func, *args, **kwargs) -> BackgroundTask:
    if not inspect.iscoroutinefunction(func):
        async def _wrapper(*a, **kw):
            return func(*a, **kw)
        return cls(_wrapper, *args, **kwargs)
    return cls(func, *args, **kwargs)

The wrapper is a simple async function that calls the sync function. No thread pool is used. The sync function runs on the event loop thread. For CPU-bound work, consider using asyncio.to_thread() instead.

10.5 Error Handling in wait()

async def wait(self, timeout: float | None = None) -> Any:
    return await self._task_obj.wait(timeout=timeout)

Delegates to Task.wait(), which:

1. Checks if result is already available (fast path)
2. Otherwise, awaits _done event with optional timeout
3. Calls _unwrap_result() which raises TaskError on failure or TaskCancelled on cancellation

---

11. Supervisor Deep Dive

11.1 State Machine

stateDiagram-v2
    [*] --> Idle : Created
    Idle --> Running : start()
    Running --> Running : Task succeeds (ALWAYS policy)
    Running --> BackingOff : Task fails
    BackingOff --> Running : delay elapsed
    Running --> Stopped : max_restarts exhausted
    Running --> Stopped : stop() called
    Running --> Stopped : Task succeeds (NEVER/ON_FAILURE)
    BackingOff --> Stopped : stop() called
    Stopped --> [*]

11.2 Restart Decision Matrix

Policy	Task Succeeds	Task Fails	Max Restarts Reached
NEVER	Stop	Stop	N/A
ALWAYS	Restart	Restart	Stop
ON_FAILURE	Stop	Restart	Stop
EXPONENTIAL_BACKOFF	Stop	Restart with delay	Stop

11.3 Backoff Calculation

delay = min(self.base_delay * (2 ** self._restarts), self.max_delay)

base_delay	max_delay	Restart 0	Restart 1	Restart 2	Restart 3	Restart 4	Restart 5
1.0	60.0	1.0s	2.0s	4.0s	8.0s	16.0s	32.0s
2.0	60.0	2.0s	4.0s	8.0s	16.0s	32.0s	60.0s
0.5	30.0	0.5s	1.0s	2.0s	4.0s	8.0s	16.0s

11.4 stop() Behavior

def stop(self) -> None:
    self._running = False
    if self._current_task:
        self._current_task.cancel()

Important: stop() sets _running = False and cancels the current task. The start() loop will exit on the next iteration. If the task is in a backoff sleep, the sleep will be interrupted by the cancellation.

11.5 wait() with Timeout

async def wait(self, timeout: float | None = None) -> None:
    await asyncio.wait_for(self._stopped.wait(), timeout=timeout)

Blocks until the supervisor stops (either by exhausting restarts or by calling stop()). Raises asyncio.TimeoutError if the timeout elapses.

11.6 Supervisor vs. BackgroundTask

Feature	BackgroundTask	Supervisor
Launch	Immediate on construction	Explicit start()
Restart	None	Configurable policy
Backoff	None	Exponential
Max restarts	None	Configurable
Monitoring	_instances class set	to_dict()
Blocking	Non-blocking	start() blocks

---

12. Circuit Breaker Deep Dive

12.1 State Diagram

stateDiagram-v2
    [*] --> CLOSED
    CLOSED --> OPEN : failures > threshold
    OPEN --> HALF_OPEN : timeout elapsed
    HALF_OPEN --> CLOSED : test request succeeds
    HALF_OPEN --> OPEN : test request fails

12.2 State Descriptions

State	Meaning	Requests
CLOSED	Normal operation	All pass through
OPEN	Failures exceeded threshold	All rejected immediately
HALF_OPEN	Testing recovery	One test request allowed

12.3 Integration Pattern

from sillo.work.types import CircuitState, CircuitBreakerOpen

class CircuitBreaker:
    def __init__(self, failure_threshold=5, recovery_timeout=30.0):
        self.state = CircuitState.CLOSED
        self.failure_count = 0
        self.failure_threshold = failure_threshold
        self.recovery_timeout = recovery_timeout
        self.last_failure_time = 0.0

    async def call(self, func, *args, **kwargs):
        if self.state == CircuitState.OPEN:
            if time.time() - self.last_failure_time > self.recovery_timeout:
                self.state = CircuitState.HALF_OPEN
            else:
                raise CircuitBreakerOpen("Circuit is open")

        try:
            result = await func(*args, **kwargs)
            if self.state == CircuitState.HALF_OPEN:
                self.state = CircuitState.CLOSED
                self.failure_count = 0
            return result
        except Exception as exc:
            self.failure_count += 1
            self.last_failure_time = time.time()
            if self.failure_count >= self.failure_threshold:
                self.state = CircuitState.OPEN
            raise

12.4 CircuitBreakerOpen Exception

class CircuitBreakerOpen(WorkError):
    """Worker circuit is open — requests are being shed."""

Inherits from WorkError and carries task_id and queue_name context.

---

13. Health and Stats Deep Dive

13.1 QueueHealth Assessment

class QueueHealth(enum.Enum):
    HEALTHY = "healthy"
    DEGRADED = "degraded"  # High backlog
    STALLED = "stalled"    # No consumers

Assessment logic:

• HEALTHY: Queue size is within normal bounds, consumers are active
• DEGRADED: Queue size is growing (backlog exceeds threshold)
• STALLED: No consumers are connected or processing

13.2 QueueStats Fields

@dataclasses.dataclass
class QueueStats:
    name: str
    size: int = 0           # Pending jobs
    completed: int = 0      # Total completed
    failed: int = 0         # Total failed
    oldest_age_ms: int = 0  # Age of oldest pending job
    status: QueueHealth = QueueHealth.HEALTHY

13.3 WorkerStats Fields

@dataclasses.dataclass
class WorkerStats:
    processed: int = 0          # Total jobs processed
    failed: int = 0             # Total jobs failed
    active: int = 0             # Currently executing
    workers: int = 0            # Number of worker coroutines
    circuit: CircuitState = CircuitState.CLOSED
    uptime_seconds: float = 0.0

13.4 SchedulerStats Fields

@dataclasses.dataclass
class SchedulerStats:
    jobs_total: int = 0
    jobs_active: int = 0
    jobs_paused: int = 0
    runs_total: int = 0
    errors_total: int = 0

13.5 TaskResult Derived Properties

@property
def duration_ms(self) -> int:
    """Wall-clock execution time in milliseconds."""
    if not self.started_at or not self.completed_at:
        return 0
    return int((self.completed_at - self.started_at) * 1000)

@property
def latency_ms(self) -> int:
    """Time from creation to start in milliseconds (queue wait)."""
    if not self.created_at or not self.started_at:
        return 0
    return int((self.started_at - self.created_at) * 1000)

@property
def ok(self) -> bool:
    return self.status == TaskStatus.COMPLETED

@property
def is_terminal(self) -> bool:
    return self.status in (TaskStatus.COMPLETED, TaskStatus.FAILED, TaskStatus.CANCELLED)

13.6 Monitoring Dashboard Pattern

async def work_health(request):
    conn = request.app.state["queue_connection"]
    sched = request.app.state.get("scheduler")

    queue_stats = await conn.size("default")
    bg_count = BackgroundTask.count()
    sched_stats = sched.stats if sched else None

    return {
        "queue": {
            "pending": queue_stats,
        },
        "background_tasks": bg_count,
        "scheduler": sched_stats.to_dict() if sched_stats else None,
    }

---

14. Exception Hierarchy Reference

classDiagram
    class WorkError {
        <<base>>
        +str task_id
        +str queue_name
    }
    class TaskError["TaskError (alias for WorkError)"]
    class TaskRejected["TaskRejected: queue refused to accept (duplicate, full)"]
    class TaskTimeout["TaskTimeout: exceeded time budget"]
    class TaskCancelled["TaskCancelled: externally cancelled"]
    class QueueFull["QueueFull: backend at capacity"]
    class BackendUnavailable["BackendUnavailable: cannot reach persistence layer"]
    class CircuitBreakerOpen["CircuitBreakerOpen: worker circuit is open"]
    class InvalidTrigger["InvalidTrigger: scheduler trigger malformed"]

    WorkError <|-- TaskError
    WorkError <|-- TaskRejected
    WorkError <|-- TaskTimeout
    WorkError <|-- TaskCancelled
    WorkError <|-- QueueFull
    WorkError <|-- BackendUnavailable
    WorkError <|-- CircuitBreakerOpen
    WorkError <|-- InvalidTrigger

14.1 Exception Context

All exceptions carry structured context:

class WorkError(Exception):
    def __init__(self, message, *, task_id="", queue_name=""):
        super().__init__(message)
        self.task_id = task_id
        self.queue_name = queue_name

This enables structured logging and error reporting:

try:
    await task.run()
except WorkError as exc:
    logger.error("Task failed: %s (task_id=%s, queue=%s)",
        exc, exc.task_id, exc.queue_name)

---

15. Integration Patterns

15.1 Request Lifecycle Integration

from sillo.work.background import BackgroundTask

async def upload_handler(request):
    file = await request.form()
    # Process file in background
    bt = BackgroundTask.run(process_upload, file.filename, file.content)
    return {"task_id": bt.id, "status": "processing"}

async def status_handler(request):
    task_id = request.path_params["task_id"]
    # Look up task by ID (would need a registry)
    return {"status": "completed"}

15.2 Supervised WebSocket Listener

from sillo.work.background import Supervisor, RestartPolicy

async def websocket_listener(url):
    async with aiohttp.ClientSession() as session:
        async with session.ws_connect(url) as ws:
            async for msg in ws:
                if msg.type == aiohttp.WSMsgType.TEXT:
                    await process_message(msg.data)
                elif msg.type == aiohttp.WSMsgType.ERROR:
                    raise ConnectionError(f"WebSocket error: {ws.exception()}")

# Supervise with exponential backoff
supervisor = Supervisor(
    websocket_listener,
    RestartPolicy.EXPONENTIAL_BACKOFF,
    max_restarts=10,
    base_delay=1.0,
    max_delay=60.0,
)

@app.on_startup
async def start_listener(app):
    asyncio.create_task(supervisor.start("wss://stream.example.com"))

@app.on_shutdown
async def stop_listener(app):
    supervisor.stop()

15.3 Background Task with Progress Tracking

from sillo.work.background import BackgroundTask
from sillo.work.types import TaskResult

class ProgressTracker:
    def __init__(self):
        self.progress = 0
        self.total = 0

    async def on_progress(self, current, total):
        self.progress = current
        self.total = total

async def process_items(items, tracker):
    for i, item in enumerate(items):
        await process_item(item)
        await tracker.on_progress(i + 1, len(items))

tracker = ProgressTracker()
bt = BackgroundTask.run(process_items, items, tracker)

15.4 Graceful Shutdown with Drain

@app.on_startup
async def startup(app):
    app.state["background_tasks"] = BackgroundTask

@app.on_shutdown
async def shutdown(app):
    result = await BackgroundTask.drain(timeout=30.0, cancel_remaining=True)
    logger.info(
        "Shutdown complete: %d total, %d completed, %d cancelled",
        result["total"], result["completed"], result["cancelled"]
    )

---

16. Performance Considerations

16.1 BackgroundTask Overhead

Each BackgroundTask creates:

• 1 Task instance (~200 bytes with slots)
• 1 asyncio.Task (~1KB)
• 1 entry in _instances set (~64 bytes)

For high-throughput scenarios (thousands of tasks per second), consider:

• Using the queue system instead of individual background tasks
• Periodically draining completed tasks to free memory
• Using NoOpHistoryManager if history is not needed

16.2 Supervisor Overhead

Each Supervisor maintains:

• 1 BackgroundTask at a time
• 1 asyncio.Event for stop signaling
• Backoff state (restart count, delay calculation)

The supervisor itself is lightweight. The main cost is the supervised task.

16.3 Memory Management

# Monitor background task count
count = BackgroundTask.count()
if count["total"] > 10000:
    logger.warning("High background task count: %d", count["total"])

# Force garbage collection of completed tasks
import gc
gc.collect()

---

17. Testing Patterns

17.1 Unit Testing BackgroundTask

import asyncio
import pytest

async def test_background_task_completes():
    async def my_task(x):
        return x * 2

    bt = BackgroundTask.run(my_task, 21)
    result = await bt.wait(timeout=5.0)
    assert result == 42
    assert bt.done
    assert bt.result.ok

async def test_background_task_failure():
    async def failing_task():
        raise ValueError("test error")

    bt = BackgroundTask.run(failing_task)
    with pytest.raises(TaskError, match="test error"):
        await bt.wait(timeout=5.0)
    assert bt.done
    assert not bt.result.ok

17.2 Unit Testing Supervisor

async def test_supervisor_restarts_on_failure():
    call_count = 0

    async def flaky_task():
        nonlocal call_count
        call_count += 1
        if call_count < 3:
            raise ValueError(f"fail #{call_count}")
        return "success"

    supervisor = Supervisor(
        flaky_task,
        RestartPolicy.ON_FAILURE,
        max_restarts=5,
        base_delay=0.01,  # Fast for testing
    )
    await supervisor.start()
    assert call_count == 3
    assert supervisor._restarts == 2

async def test_supervisor_exhausts_restarts():
    async def always_fail():
        raise ValueError("always fails")

    supervisor = Supervisor(
        always_fail,
        RestartPolicy.EXPONENTIAL_BACKOFF,
        max_restarts=3,
        base_delay=0.01,
    )
    await supervisor.start()
    assert supervisor._restarts == 3

17.3 Testing Drain

async def test_drain_waits_for_all():
    async def slow_task():
        await asyncio.sleep(0.1)

    tasks = [BackgroundTask.run(slow_task) for _ in range(5)]
    result = await BackgroundTask.drain(timeout=5.0, cancel_remaining=False)
    assert result["total"] == 5
    assert result["completed"] == 5
    assert result["cancelled"] == 0

---

18. Common Pitfalls

18.1 Forgetting to Await wait()

# Wrong — task may not be done when you check
bt = BackgroundTask.run(some_task)
print(bt.done)  # May be False

# Right — wait for completion
bt = BackgroundTask.run(some_task)
await bt.wait()
print(bt.done)  # Always True

18.2 Using BackgroundTask for CPU-Bound Work

# Wrong — blocks the event loop
bt = BackgroundTask.run(heavy_computation)

# Right — use thread pool
import asyncio
bt = BackgroundTask.run_sync(lambda: asyncio.to_thread(heavy_computation))

18.3 Not Handling Supervisor Stop

# Wrong — supervisor keeps running after app shutdown
supervisor = Supervisor(my_task, RestartPolicy.ALWAYS)
asyncio.create_task(supervisor.start())

# Right — wire into app lifecycle
@app.on_shutdown
async def shutdown():
    supervisor.stop()
    await supervisor.wait(timeout=10.0)

18.4 Assuming drain() Cancels Immediately

# drain() waits up to timeout, THEN cancels remaining
result = await BackgroundTask.drain(timeout=5.0, cancel_remaining=True)
# Tasks may have completed during the 5-second wait