radical.asyncflow.backends.execution.radical_pilot¶

def __await__(self):
    """Make RadicalExecutionBackend awaitable."""
    return self._async_init().__await__()

def get_task_states_map(self):
    """Get the state mapper for this backend.

    Returns:
        StateMapper: StateMapper instance configured for RadicalPilot backend
            with appropriate state mappings (DONE, FAILED, CANCELED, AGENT_EXECUTING).
    """
    return StateMapper(backend=self)

def setup_raptor_mode(self, raptor_config):
    """Set up Raptor mode by configuring and submitting master and worker tasks.

    Initializes Raptor mode by creating master tasks and their associated
    worker tasks based on the provided configuration. Masters coordinate
    work distribution while workers execute the actual computations.

    Args:
        raptor_config (Dict): Configuration dictionary with the following structure:
            {
                'masters': [
                    {
                        'executable': str,  # Path to master executable
                        'arguments': list,  # Arguments for master
                        'ranks': int,       # Number of CPU processes
                        'workers': [        # Worker configurations
                            {
                                'executable': str,    # Worker executable path
                                'arguments': list,    # Worker arguments
                                'ranks': int,         # Worker CPU processes
                                'worker_type': str    # Optional worker class
                            },
                            ...
                        ]
                    },
                    ...
                ]
            }

    Raises:
        Exception: If task description creation or submission fails.

    Note:
        - Creates unique UIDs for masters and workers using session namespace
        - Sets up master selector for load balancing across masters
        - Workers default to 'DefaultWorker' class if not specified
        - All master and worker tasks are stored in respective class attributes
    """

    self.masters = []
    self.workers = []
    self.master_selector = self.select_master()

    cfg = copy.deepcopy(raptor_config)
    masters = cfg['masters']

    for master_description in masters:
        workers = master_description.pop('workers')

        md = rp.TaskDescription(master_description)
        md.uid = ru.generate_id('flow.master.%(item_counter)06d', ru.ID_CUSTOM,
                                 ns=self.session.uid)
        md.mode = rp.RAPTOR_MASTER
        master = self.resource_pilot.submit_raptors(md)[0]
        self.masters.append(master)

        for worker_description in workers:
            # Set default worker class and override if specified
            raptor_class = worker_description.pop('worker_type', 'DefaultWorker')

            # Create and configure worker
            worker = master.submit_workers(
                rp.TaskDescription({
                    **worker_description,
                    'raptor_id': md.uid,
                    'mode': rp.RAPTOR_WORKER,
                    'raptor_class': raptor_class,
                    'uid': ru.generate_id('flow.worker.%(item_counter)06d', 
                                          ru.ID_CUSTOM, ns=self.session.uid)}))

            self.workers.append(worker)

def select_master(self):
    """Create a generator for load balancing task submission across masters.

    Provides a round-robin generator that cycles through available master
    UIDs to distribute tasks evenly across all masters in Raptor mode.

    Returns:
        Generator[str]: Generator yielding master UIDs in round-robin fashion.

    Raises:
        RuntimeError: If Raptor mode is not enabled or no masters are available.

    Example:
        ::
            selector = backend.select_master()
            master_uid = next(selector)  # Get next master for task assignment
    """
    if not self.raptor_mode or not self.masters:
        raise RuntimeError('Raptor mode is not enabled or no masters available')

    current_master = 0
    masters_uids = [m.uid for m in self.masters]

    while True:
        yield masters_uids[current_master]
        current_master = (current_master + 1) % len(self.masters)

def register_callback(self, func):
    """Register a callback function for task state changes.

    Sets up a callback mechanism that handles task state transitions,
    with special handling for service tasks that require additional
    readiness confirmation.

    Args:
        func (Callable): Callback function that will be invoked on task state changes.
            Should accept parameters: (task, state, service_callback=None).

    Note:
        - Service tasks in AGENT_EXECUTING state get special service_ready_callback
        - All other tasks use the standard callback mechanism
        - The callback is registered with the underlying task manager
    """

    self._callback_func = func

    def backend_callback(task, state):
        service_callback = None
        # Attach backend-specific done_callback for service tasks
        if task.mode == rp.TASK_SERVICE and state == rp.AGENT_EXECUTING:
            service_callback = service_ready_callback

        # Forward to workflow manager's standard callback
        func(task, state, service_callback=service_callback)

    self.task_manager.register_callback(backend_callback)

def build_task(self, uid, task_desc, task_backend_specific_kwargs) -> rp.TaskDescription:
    """Build a RadicalPilot task description from workflow task parameters.

    Converts a workflow task description into a RadicalPilot TaskDescription,
    handling different task modes (executable, function, service) and applying
    appropriate configurations.

    Args:
        uid (str): Unique identifier for the task.
        task_desc (Dict): Task description containing:
            - executable: Path to executable (for executable tasks)
            - function: Python function (for function tasks)
            - args: Function arguments
            - kwargs: Function keyword arguments
            - is_service: Boolean indicating service task
        task_backend_specific_kwargs (Dict): RadicalPilot-specific parameters
            for the task description.

    Returns:
        rp.TaskDescription: Configured RadicalPilot task description, or None
            if task creation failed.

    Note:
        - Function tasks require Raptor mode to be enabled
        - Service tasks cannot be Python functions
        - Failed tasks trigger callback with FAILED state
        - Raptor tasks are assigned to masters via load balancing

    Example:
        ::
            task_desc = {
                'executable': '/bin/echo',
                'args': ['Hello World'],
                'is_service': False
            }
            rp_task = backend.build_task('task_001', task_desc, {})
    """

    is_service = task_desc.get('is_service', False)
    rp_task = rp.TaskDescription(from_dict=task_backend_specific_kwargs)
    rp_task.uid = uid

    if task_desc['executable']:
        rp_task.mode = rp.TASK_SERVICE if is_service else rp.TASK_EXECUTABLE
        rp_task.executable = task_desc['executable']
    elif task_desc['function']:
        if is_service:
            error_msg = 'RadicalExecutionBackend does not support function service tasks'
            rp_task['exception'] = ValueError(error_msg)
            self._callback_func(rp_task, rp.FAILED)
            return

        rp_task.mode = rp.TASK_FUNCTION
        rp_task.function = rp.PythonTask(task_desc['function'],
                                         task_desc['args'],
                                         task_desc['kwargs'])

    if rp_task.mode in [rp.TASK_FUNCTION, rp.TASK_EVAL,
                        rp.TASK_PROC, rp.TASK_METHOD]:
        if not self.raptor_mode:
            error_msg = f'Raptor mode is not enabled, cannot register {rp_task.mode}'
            rp_task['exception'] =  RuntimeError(error_msg)
            self._callback_func(rp_task, rp.FAILED)
            return

        rp_task.raptor_id = next(self.master_selector)

    self.tasks[uid] = rp_task

    return rp_task

def link_explicit_data_deps(self, src_task=None, dst_task=None, file_name=None, file_path=None):
    """Link explicit data dependencies between tasks or from external sources.

    Creates data staging entries to establish explicit dependencies where
    files are transferred or linked from source to destination tasks.
    Supports both task-to-task dependencies and external file staging.

    Args:
        src_task (Optional[Dict]): Source task dictionary containing the file.
            None when staging from external path.
        dst_task (Dict): Destination task dictionary that will receive the file.
            Must contain 'task_backend_specific_kwargs' key.
        file_name (Optional[str]): Name of the file to stage. Defaults to:
            - src_task UID if staging from task
            - basename of file_path if staging from external path
        file_path (Optional[str]): External file path to stage (alternative
            to task-sourced files).

    Returns:
        Dict: The data dependency dictionary that was added to input staging.

    Raises:
        ValueError: If neither file_name nor file_path is provided, or if
            src_task is missing when file_path is not specified.

    Note:
        - External files use TRANSFER action
        - Task-to-task dependencies use LINK action
        - Files are staged to task:/// namespace in destination
        - Input staging list is created if it doesn't exist

    Example:
        ::
            # Link output from task1 to task2
            backend.link_explicit_data_deps(
                src_task={'uid': 'task1'},
                dst_task={'task_backend_specific_kwargs': {}},
                file_name='output.dat'
            )

            # Stage external file
            backend.link_explicit_data_deps(
                dst_task={'task_backend_specific_kwargs': {}},
                file_path='/path/to/input.txt'
            )
    """
    if not file_name and not file_path:
        raise ValueError('Either file_name or file_path must be provided')

    dst_kwargs = dst_task['task_backend_specific_kwargs']

    # Determine the filename if not provided
    if not file_name:
        if file_path:
            file_name = file_path.split('/')[-1]  # Use basename from path
        elif src_task:
            file_name = src_task['uid']  # Fallback to task UID
        else:
            raise ValueError("Must provide either file_name, file_path, or src_task")

    # Create the appropriate data dependency
    if file_path:
        data_dep = {
            'source': file_path,
            'target': f"task:///{file_name}",
            'action': rp.TRANSFER}
    else:
        if not src_task:
            raise ValueError("src_task must be provided when file_path is not specified")
        data_dep = {
            'source': f"pilot:///{src_task['uid']}/{file_name}",
            'target': f"task:///{file_name}",
            'action': rp.LINK}

    # Add to input staging
    if 'input_staging' not in dst_kwargs:
        dst_kwargs['input_staging'] = [data_dep]
    else:
        dst_kwargs['input_staging'].append(data_dep)

    return data_dep

def link_implicit_data_deps(self, src_task, dst_task):
    """Add implicit data dependencies through symbolic links in task sandboxes.

    Creates pre-execution commands that establish symbolic links from the
    source task's sandbox to the destination task's sandbox, simulating
    implicit data dependencies without explicit file specifications.

    Args:
        src_task (Dict): Source task dictionary containing 'uid' key.
        dst_task (Dict): Destination task dictionary with 'task_backend_specific_kwargs'.

    Note:
        - Links all files from source sandbox except the task UID file itself
        - Uses environment variables for source task identification
        - Commands are added to the destination task's pre_exec list
        - Symbolic links are created in the destination task's sandbox

    Implementation Details:
        1. Sets SRC_TASK_ID environment variable
        2. Sets SRC_TASK_SANDBOX path variable
        3. Creates symbolic links for all files except the task ID file

    Example:
        ::
            src_task = {'uid': 'producer_task'}
            dst_task = {'task_backend_specific_kwargs': {}}
            backend.link_implicit_data_deps(src_task, dst_task)
    """

    dst_kwargs = dst_task['task_backend_specific_kwargs']
    src_uid = src_task['uid']

    cmd1 = f'export SRC_TASK_ID={src_uid}'
    cmd2 = f'export SRC_TASK_SANDBOX="$RP_PILOT_SANDBOX/$SRC_TASK_ID"'

    cmd3 = '''files=$(cd "$SRC_TASK_SANDBOX" && ls | grep -ve "^$SRC_TASK_ID")
            for f in $files
            do 
                ln -sf "$SRC_TASK_SANDBOX/$f" "$RP_TASK_SANDBOX"
            done'''

    commands = [cmd1, cmd2, cmd3]

    if dst_kwargs.get('pre_exec'):
        dst_kwargs['pre_exec'].extend(commands)
    else:
        dst_kwargs['pre_exec'] = commands

async def submit_tasks(self, tasks: list):
    """Submit a list of tasks for execution.

    Processes a list of workflow tasks, builds RadicalPilot task descriptions,
    and submits them to the task manager for execution. Handles task building
    failures gracefully by skipping invalid tasks.

    Args:
        tasks (list): List of task dictionaries, each containing:
            - uid: Unique task identifier
            - task_backend_specific_kwargs: RadicalPilot-specific parameters
            - Other task description fields

    Returns:
        The result of task_manager.submit_tasks() with successfully built tasks.

    Note:
        - Failed task builds are skipped (build_task returns None)
        - Only successfully built tasks are submitted to the task manager
        - Task building includes validation and error handling
    """

    _tasks = []
    for task in tasks:
        task_to_submit = self.build_task(task['uid'],
                      task, task['task_backend_specific_kwargs'])

        if not task_to_submit:
            continue

        _tasks.append(task_to_submit)

    return self.task_manager.submit_tasks(_tasks)

async def cancel_task(self, uid: str) -> bool:
    """
    Cancel a task in the execution backend.

    Args:
        uid (str): The UID of the task to cancel.

    Returns:
        bool: True if the task was found and cancellation was attempted, False otherwise.
    """
    if uid in self.tasks:
        self.task_manager.cancel_tasks(uid)
        return True
    return False

def get_nodelist(self):
    """Get information about allocated compute nodes.

    Retrieves the nodelist from the active resource pilot, providing
    details about the compute nodes allocated for task execution.

    Returns:
        rp.NodeList: NodeList object containing information about allocated
            nodes. Each node in nodelist.nodes is of type rp.NodeResource.
            Returns None if the pilot is not in PMGR_ACTIVE state.

    Note:
        - Only returns nodelist when pilot is in active state
        - Nodelist provides detailed resource information for each node
        - Useful for resource-aware task scheduling and monitoring
    """
    nodelist = None
    if self.resource_pilot.state == rp.PMGR_ACTIVE:
        nodelist = self.resource_pilot.nodelist

    return nodelist

def state(self):
    """Retrieve the current state of the resource pilot.

    Returns:
        The current state of the resource pilot.

    Note:
        This method is currently not implemented and serves as a placeholder.
    """
    raise NotImplementedError

def task_state_cb(self, task, state):
    """Callback function for handling task state changes.

    Args:
        task: The task object whose state has changed.
        state: The new state of the task.

    Note:
        This method is currently not implemented and serves as a placeholder
        for custom task state change handling.
    """
    raise NotImplementedError

async def shutdown(self) -> None:
    """Gracefully shutdown the backend and clean up resources.

    Closes the RadicalPilot session with data download, ensuring proper
    cleanup of all resources including pilots, tasks, and session data.

    Note:
        - Downloads session data before closing
        - Ensures graceful termination of all backend resources
        - Prints confirmation message when shutdown is triggered
    """
    print('Shutdown is triggered, terminating the resources gracefully')
    self.session.close(download=True)

async def __aenter__(self):
    """Async context manager entry."""
    if not self._initialized:
        await self._async_init()
    return self

async def __aexit__(self, exc_type, exc_val, exc_tb):
    """Async context manager exit."""
    await self.shutdown()

@classmethod
async def create(cls, resources: Dict, raptor_config: Optional[Dict] = None):
    """Alternative factory method for creating initialized backend."""
    backend = cls(resources, raptor_config)
    return await backend