import logging
import os
import queue
import signal
import threading
import time
from typing import Any, Callable, Dict, Generator, Iterable, List, Optional, Sized, Union, Tuple
try:
import numpy as np
NUMPY_INSTALLED = True
except ImportError:
np = None
NUMPY_INSTALLED = False
from mpire.async_result import (AsyncResult, AsyncResultType, AsyncResultWithExceptionGetter,
UnorderedAsyncExitResultIterator, UnorderedAsyncResultIterator)
from mpire.comms import EXIT_FUNC, INIT_FUNC, MAIN_PROCESS, POISON_PILL, WorkerComms
from mpire.context import DEFAULT_START_METHOD, RUNNING_WINDOWS
from mpire.dashboard.connection_utils import get_dashboard_connection_details
from mpire.exception import populate_exception
from mpire.insights import WorkerInsights
from mpire.params import check_map_parameters, CPUList, WorkerMapParams, WorkerPoolParams
from mpire.progress_bar import ProgressBarHandler
from mpire.signal import DisableKeyboardInterruptSignal
from mpire.tqdm_utils import get_tqdm, TqdmManager
from mpire.utils import apply_numpy_chunking, chunk_tasks, set_cpu_affinity
from mpire.worker import MP_CONTEXTS, worker_factory
logger = logging.getLogger(__name__)
[docs]class WorkerPool:
"""
A multiprocessing worker pool which acts like a ``multiprocessing.Pool``, but is faster and has more options.
"""
[docs] def __init__(self, n_jobs: Optional[int] = None, daemon: bool = True, cpu_ids: CPUList = None,
shared_objects: Any = None, pass_worker_id: bool = False, use_worker_state: bool = False,
start_method: str = DEFAULT_START_METHOD, keep_alive: bool = False, use_dill: bool = False,
enable_insights: bool = False, order_tasks: bool = False) -> None:
"""
:param n_jobs: Number of workers to spawn. If ``None``, will use ``mpire.cpu_count()``
:param daemon: Whether to start the child processes as daemon
:param cpu_ids: List of CPU IDs to use for pinning child processes to specific CPUs. The list must be as long as
the number of jobs used (if ``n_jobs`` equals ``None`` it must be equal to ``mpire.cpu_count()``), or the
list must have exactly one element. In the former case, element `i` specifies the CPU ID(s) to use for child
process `i`. In the latter case the single element specifies the CPU ID(s) for all child processes to use.
A single element can be either a single integer specifying a single CPU ID, or a list of integers specifying
that a single child process can make use of multiple CPU IDs. If ``None``, CPU pinning will be disabled
:param shared_objects: Objects to be passed on as shared objects to the workers once. It will be passed on to
the target, ``worker_init``, and ``worker_exit`` functions. ``shared_objects`` is only passed on when it's
not ``None``. Shared objects will be copy-on-write when using ``fork`` as start method. When enabled,
functions receive the shared objects as second argument, depending on other settings. The order is:
``worker_id``, ``shared_objects``, ``worker_state``, and finally the arguments passed on from
``iterable_of_args``
:param pass_worker_id: Whether to pass on a worker ID to the target, ``worker_init``, and ``worker_exit``
functions. When enabled, functions receive the worker ID as first argument, depending on other settings. The
order is: ``worker_id``, ``shared_objects``, ``worker_state``, and finally the arguments passed on from
``iterable_of_args``
:param use_worker_state: Whether to let a worker have a worker state. The worker state will be passed on to the
target, ``worker_init``, and ``worker_exit`` functions. When enabled, functions receive the worker state as
third argument, depending on other settings. The order is: ``worker_id``, ``shared_objects``,
``worker_state``, and finally the arguments passed on from ``iterable_of_args``
:param start_method: Which process start method to use. Options for multiprocessing: ``'fork'`` (default, if
available), ``'forkserver'`` and ``'spawn'`` (default, if ``'fork'`` isn't available). For multithreading
use ``'threading'``. See https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods
for more information and
https://docs.python.org/3/library/multiprocessing.html#the-spawn-and-forkserver-start-methods for some
caveats when using the ``'spawn'`` or ``'forkserver'`` methods
:param keep_alive: When ``True`` it will keep workers alive after completing a map call, allowing to reuse
workers
:param use_dill: Whether to use dill as serialization backend. Some exotic types (e.g., lambdas, nested
functions) don't work well when using ``spawn`` as start method. In such cased, use ``dill`` (can be a bit
slower sometimes)
:param enable_insights: Whether to enable worker insights. Might come at a small performance penalty (often
neglible)
:param order_tasks: Whether to provide tasks to the workers in order, such that worker 0 will get chunk 0,
worker 1 will get chunk 1, etc.
"""
# Set parameters
self.pool_params = WorkerPoolParams(n_jobs, cpu_ids, daemon, shared_objects, pass_worker_id, use_worker_state,
start_method, keep_alive, use_dill, enable_insights, order_tasks)
self.map_params = None # type: Optional[WorkerMapParams]
# Worker factory
self.Worker = worker_factory(start_method, use_dill)
# Multiprocessing context
if start_method == 'threading':
self.ctx = MP_CONTEXTS['threading']
else:
self.ctx = MP_CONTEXTS['mp_dill' if use_dill else 'mp'][start_method]
# Cache for storing intermediate results. Add result objects for the worker_init and worker_exit functions
self._cache: Dict[int, AsyncResultType] = {}
AsyncResultWithExceptionGetter(self._cache, MAIN_PROCESS)
AsyncResultWithExceptionGetter(self._cache, INIT_FUNC)
UnorderedAsyncExitResultIterator(self._cache)
# Container of the child processes and corresponding communication objects
self._workers = []
self._worker_comms = WorkerComms(self.ctx, self.pool_params.n_jobs, self.pool_params.order_tasks)
self._map_running = False
# Threads needed for gathering results, restarts, and checking for unexpective deaths and timeouts
self._results_handler_thread = None
self._restart_handler_thread = None
self._timeout_handler_thread = None
self._unexpected_death_handler_thread = None
self._handler_threads_stop_event = threading.Event()
# Progress bar handler, in case it is used
self._progress_bar_handler = None
# Worker insights, used for profiling
self._worker_insights = WorkerInsights(self.ctx, self.pool_params.n_jobs, self.pool_params.use_dill)
[docs] def pass_on_worker_id(self, pass_on: bool = True) -> None:
"""
Set whether to pass on the worker ID to the function to be executed or not (default= ``False``).
:param pass_on: Whether to pass on a worker ID to the target, ``worker_init``, and ``worker_exit``
functions. When enabled, functions receive the worker ID depending on other settings. The order is:
``worker_id``, ``shared_objects``, ``worker_state``, and finally the arguments passed on using
``iterable_of_args``
"""
if pass_on != self.pool_params.pass_worker_id:
self._worker_comms.reset()
self.pool_params.pass_worker_id = pass_on
[docs] def set_shared_objects(self, shared_objects: Any = None) -> None:
"""
Set shared objects to pass to the workers.
:param shared_objects: Objects to be passed on as shared objects to the workers once. It will be passed on to
the target, ``worker_init``, and ``worker_exit`` functions. ``shared_objects`` is only passed on when it's
not ``None``. Shared objects will be copy-on-write when using ``fork`` as start method. When enabled,
functions receive the shared objects depending on other settings. The order is: ``worker_id``,
``shared_objects``, ``worker_state``, and finally the arguments passed on using ``iterable_of_args```
"""
if shared_objects != self.pool_params.shared_objects:
self._worker_comms.reset()
self.pool_params.shared_objects = shared_objects
[docs] def set_use_worker_state(self, use_worker_state: bool = True) -> None:
"""
Set whether or not each worker should have its own state variable. Each worker has its own state, so it's not
shared between the workers.
:param use_worker_state: Whether to let a worker have a worker state. The worker state will be passed on to the
target, ``worker_init``, and ``worker_exit`` functions. When enabled, functions receive the worker state
depending on other settings. The order is: ``worker_id``, ``shared_objects``, ``worker_state``, and finally
the arguments passed on using ``iterable_of_args``
"""
if use_worker_state != self.pool_params.use_worker_state:
self._worker_comms.reset()
self.pool_params.use_worker_state = use_worker_state
[docs] def set_keep_alive(self, keep_alive: bool = True) -> None:
"""
Set whether workers should be kept alive in between consecutive map calls.
:param keep_alive: When True it will keep workers alive after completing a map call, allowing to reuse workers
"""
self.pool_params.keep_alive = keep_alive
[docs] def set_order_tasks(self, order_tasks: bool = True) -> None:
"""
Set whether to provide tasks to the workers in order, such that worker 0 will get chunk 0, worker 1 will get
chunk 1, etc.
:param order_tasks: Whether to provide tasks to the workers in order, such that worker 0 will get chunk 0,
worker 1 will get chunk 1, etc.
"""
self.pool_params.order_tasks = order_tasks
def _start_workers(self) -> None:
"""
Spawns the workers and starts them so they're ready to start reading from the tasks queue.
"""
self._cache[MAIN_PROCESS].reset()
self._cache[INIT_FUNC].reset()
self._cache[EXIT_FUNC].reset()
# Init communication primitives
self._worker_comms.init_comms()
self._worker_insights.reset_insights(self.pool_params.enable_insights)
# Start new workers
self._workers = [None] * self.pool_params.n_jobs
for worker_id in range(self.pool_params.n_jobs):
self._start_worker(worker_id)
# Start results listener, restart handler, timeout handler, unexpected death handler
self._handler_threads_stop_event.clear()
self._results_handler_thread = threading.Thread(target=self._results_handler, daemon=True)
self._restart_handler_thread = threading.Thread(target=self._restart_handler, daemon=True)
self._timeout_handler_thread = threading.Thread(target=self._timeout_handler, daemon=True)
self._unexpected_death_handler_thread = threading.Thread(target=self._unexpected_death_handler, daemon=True)
self._results_handler_thread.start()
self._restart_handler_thread.start()
self._timeout_handler_thread.start()
self._unexpected_death_handler_thread.start()
def _start_worker(self, worker_id: int) -> None:
"""
Creates and starts a single worker
:param worker_id: ID of the worker
:return: Worker instance
"""
# Disable the interrupt signal. We let the process die gracefully if it needs to
with DisableKeyboardInterruptSignal():
# Create worker
self._workers[worker_id] = self.Worker(
worker_id, self.pool_params, self.map_params, self._worker_comms, self._worker_insights,
TqdmManager.get_connection_details(), get_dashboard_connection_details(), time.time()
)
self._workers[worker_id].daemon = self.pool_params.daemon
self._workers[worker_id].name = f"Worker-{worker_id}"
self._workers[worker_id].start()
# Pin CPU if desired
if self.pool_params.cpu_ids:
set_cpu_affinity(self._workers[worker_id].pid, self.pool_params.cpu_ids[worker_id])
def _results_handler(self) -> None:
"""
Listen for results from the workers and add it to the cache. Note that when ``set`` is called on a result
object, the result is automatically removed from the cache.
"""
while True:
results_batch = self._worker_comms.get_results(block=True)
for job_id, success, result in results_batch:
# Poison pill, stop the listener
if isinstance(result, str) and result == POISON_PILL:
return
try:
if success:
self._cache[job_id]._set(success=True, result=result)
else:
err, traceback_err = populate_exception(*result)
err.__cause__ = traceback_err
# When a worker_init times out, the pool shuts down and we set all tasks that haven't completed
# yet to failed
job_ids = ((set(self._cache.keys()) - {MAIN_PROCESS, EXIT_FUNC}) if job_id == INIT_FUNC else
{job_id})
for _job_id in job_ids:
self._cache[_job_id]._set(success=False, result=err)
except KeyError:
# This can happen if the job has already been removed from the cache, which can occur if the job
# has been cancelled, or if the job has been removed from the cache because the timeout has
# expired
pass
def _restart_handler(self) -> None:
"""
Listen for worker restarts and restart them if needed.
"""
while not self._worker_comms.exception_thrown() and not self._handler_threads_stop_event.is_set():
for worker_id in self._worker_comms.get_worker_restarts():
# The get_worker_restarts call is blocking, but can unblock when we need to stop (either due to an
# exception or because all tasks have been processed). However, a worker could've asked for a restart in
# the meantime, so we need to check if we need to stop again
if self._worker_comms.exception_thrown() or self._handler_threads_stop_event.is_set():
return
# Join worker. This can take a while as the worker could still be holding on to data it needs to send
# over the results queue
try:
self._workers[worker_id].join()
except OSError:
# This can happen if the worker has already died (e.g., killed by the OS)
pass
# Start new worker
self._worker_comms.reset_worker_restart(worker_id)
self._start_worker(worker_id)
def _unexpected_death_handler(self) -> None:
"""
Checks that workers that are supposed to be alive, are actually alive. If not, then a worker died unexpectedly.
Terminate signals are handled by workers themselves, but if a worker dies for any other reason, then we need
to handle it here.
Note that a worker can be alive, but their alive status is still False. This doesn't really matter, because we
know the worker is alive according to the OS. The only way we know that something bad happened is when a worker
is supposed to be alive but according to the OS it's not.
"""
while not self._worker_comms.exception_thrown() and not self._handler_threads_stop_event.is_set():
# This thread can be started before the workers are created, so we need to check that they exist. If not
# we just wait a bit and try again.
for worker_id in range(len(self._workers)):
try:
worker_died = (self._worker_comms.is_worker_alive(worker_id) and
not self._workers[worker_id].is_alive())
except ValueError:
worker_died = False
if worker_died:
# Obtain task it was working on and set it to failed
job_id = self._worker_comms.get_worker_working_on_job(worker_id)
self._worker_comms.signal_exception_thrown(job_id)
err = RuntimeError(f"Worker-{worker_id} died unexpectedly")
# When a worker dies unexpectedly, the pool shuts down and we set all tasks that haven't completed
# yet to failed
job_ids = set(self._cache.keys()) - {MAIN_PROCESS}
for job_id in job_ids:
self._cache[job_id]._set(success=False, result=err)
return
# Check this every once in a while
time.sleep(0.1)
def _timeout_handler(self) -> None:
"""
Check for worker_init/task/worker_exit timeouts
"""
def _get_init_config() -> Tuple[str, Optional[float], Callable[[int, float], bool]]:
return 'worker_init', self.map_params.worker_init_timeout, self._worker_comms.has_worker_init_timed_out
def _get_exit_config() -> Tuple[str, Optional[float], Callable[[int, float], bool]]:
return 'worker_exit', self.map_params.worker_exit_timeout, self._worker_comms.has_worker_exit_timed_out
def _get_task_config(_job_id) -> Tuple[str, Optional[float], Callable[[int, float], bool]]:
try:
return 'task', self._cache[job_id]._timeout, self._worker_comms.has_worker_task_timed_out
except KeyError:
return 'task', None, self._worker_comms.has_worker_task_timed_out
while not self._worker_comms.exception_thrown() and not self._handler_threads_stop_event.is_set():
# We're making a shallow copy here to avoid dictionary changes size during iteration errors
if (
self.map_params.worker_init_timeout is None
and self.map_params.worker_exit_timeout is None
and all(job._timeout is None for job in self._cache.copy().values())
):
# No timeouts set, so no need to check
time.sleep(0.1)
continue
for worker_id in range(self.pool_params.n_jobs):
# Obtain what the worker is working on and obtain corresponding timeout setting
job_id = self._worker_comms.get_worker_working_on_job(worker_id)
if job_id == INIT_FUNC:
timeout_func_name, timeout_var, has_timed_out_func = _get_init_config()
elif job_id == EXIT_FUNC:
timeout_func_name, timeout_var, has_timed_out_func = _get_exit_config()
else:
timeout_func_name, timeout_var, has_timed_out_func = _get_task_config(job_id)
# If timeout has expired set job to failed
if timeout_var is not None and has_timed_out_func(worker_id, timeout_var):
# If we're dealing with a map/init/exit task, send a kill signal to all workers. Otherwise, we're
# dealing with an apply task and we only interrupt that one
kill_pool = (
job_id in {INIT_FUNC, EXIT_FUNC} or
isinstance(self._cache[job_id], UnorderedAsyncResultIterator)
)
if kill_pool:
self._worker_comms.signal_exception_thrown(job_id)
self._send_kill_signal_to_worker(worker_id)
# When a worker_init times out, the pool shuts down and we set all tasks that haven't completed yet
# to failed
err = TimeoutError(f"Worker-{worker_id} {timeout_func_name} timed out (timeout={timeout_var})")
job_ids = (set(self._cache.keys()) - {MAIN_PROCESS, EXIT_FUNC}) if job_id == INIT_FUNC else {job_id}
for job_id in job_ids:
self._cache[job_id]._set(success=False, result=err)
if kill_pool:
return
# Check this every once in a while
time.sleep(0.1)
[docs] def get_exit_results(self) -> List:
"""
Obtain a list of exit results when an exit function is defined.
:return: Exit results list
"""
return self._cache[EXIT_FUNC].get_results()
[docs] def __enter__(self) -> 'WorkerPool':
"""
Enable the use of the ``with`` statement.
"""
return self
[docs] def __exit__(self, *_: Any) -> None:
"""
Enable the use of the ``with`` statement. Gracefully terminates workers, if there are any
"""
self.terminate()
[docs] def map(self, func: Callable, iterable_of_args: Union[Sized, Iterable], iterable_len: Optional[int] = None,
max_tasks_active: Optional[int] = None, chunk_size: Optional[int] = None, n_splits: Optional[int] = None,
worker_lifespan: Optional[int] = None, progress_bar: bool = False, concatenate_numpy_output: bool = True,
worker_init: Optional[Callable] = None, worker_exit: Optional[Callable] = None,
task_timeout: Optional[float] = None, worker_init_timeout: Optional[float] = None,
worker_exit_timeout: Optional[float] = None, progress_bar_options: Optional[Dict[str, Any]] = None,
progress_bar_style: Optional[str] = None) -> Any:
"""
Same as ``multiprocessing.map()``. Also allows a user to set the maximum number of tasks available in the queue.
Note that this function can be slower than the unordered version.
:param func: Function to call each time new task arguments become available. When passing on the worker ID the
function should receive the worker ID as its first argument. If shared objects are provided the function
should receive those as the next argument. If the worker state has been enabled it should receive a state
variable as the next argument
:param iterable_of_args: A numpy array or an iterable containing tuples of arguments to pass to a worker, which
passes it to the function ``func``
:param iterable_len: Number of elements in the ``iterable_of_args``. When chunk_size is set to ``None`` it needs
to know the number of tasks. This can either be provided by implementing the ``__len__`` function on the
iterable object, or by specifying the number of tasks
:param max_tasks_active: Maximum number of active tasks in the queue. If ``None`` it will be converted to
``n_jobs * chunk_size * 2``
:param chunk_size: Number of simultaneous tasks to give to a worker. When ``None`` it will use ``n_splits``.
:param n_splits: Number of splits to use when ``chunk_size`` is ``None``. When both ``chunk_size`` and
``n_splits`` are ``None``, it will use ``n_splits = n_jobs * 64``.
:param worker_lifespan: Number of tasks a worker can handle before it is restarted. If ``None``, workers will
stay alive the entire time. Use this when workers use up too much memory over the course of time
:param progress_bar: When ``True`` it will display a progress bar
:param concatenate_numpy_output: When ``True`` it will concatenate numpy output to a single numpy array
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param progress_bar_options: Dictionary containing keyword arguments to pass to the ``tqdm`` progress bar. See
``tqdm.tqdm()`` for details. The arguments ``total`` and ``leave`` will be overwritten by MPIRE.
:param progress_bar_style: The progress bar style to use. Can be one of ``None``, ``'std'``, or ``'notebook'``
:return: List with ordered results
"""
# Notify workers to keep order in mind
self._worker_comms.signal_keep_order()
# If we're dealing with numpy arrays, we have to chunk them here already
if NUMPY_INSTALLED and isinstance(iterable_of_args, np.ndarray):
iterable_of_args, iterable_len, chunk_size, n_splits = apply_numpy_chunking(iterable_of_args, iterable_len,
chunk_size, n_splits,
self.pool_params.n_jobs)
# Process all args
if iterable_len is None and hasattr(iterable_of_args, '__len__'):
iterable_len = len(iterable_of_args)
results = self.map_unordered(
func, ((args_idx, args) for args_idx, args in enumerate(iterable_of_args)), iterable_len, max_tasks_active,
chunk_size, n_splits, worker_lifespan, progress_bar, worker_init, worker_exit, task_timeout,
worker_init_timeout, worker_exit_timeout, progress_bar_options, progress_bar_style
)
# Notify workers to forget about order
self._worker_comms.clear_keep_order()
# Rearrange and return
sorted_results = [result[1] for result in sorted(results, key=lambda result: result[0])]
# Convert back to numpy if necessary
return (np.concatenate(sorted_results) if NUMPY_INSTALLED and sorted_results and concatenate_numpy_output and
isinstance(sorted_results[0], np.ndarray) else sorted_results)
[docs] def map_unordered(self, func: Callable, iterable_of_args: Union[Sized, Iterable],
iterable_len: Optional[int] = None, max_tasks_active: Optional[int] = None,
chunk_size: Optional[int] = None, n_splits: Optional[int] = None,
worker_lifespan: Optional[int] = None, progress_bar: bool = False,
worker_init: Optional[Callable] = None, worker_exit: Optional[Callable] = None,
task_timeout: Optional[float] = None, worker_init_timeout: Optional[float] = None,
worker_exit_timeout: Optional[float] = None,
progress_bar_options: Optional[Dict[str, Any]] = None,
progress_bar_style: Optional[str] = None) -> Any:
"""
Same as ``multiprocessing.map()``, but unordered. Also allows a user to set the maximum number of tasks
available in the queue.
:param func: Function to call each time new task arguments become available. When passing on the worker ID the
function should receive the worker ID as its first argument. If shared objects are provided the function
should receive those as the next argument. If the worker state has been enabled it should receive a state
variable as the next argument
:param iterable_of_args: A numpy array or an iterable containing tuples of arguments to pass to a worker, which
passes it to the function ``func``
:param iterable_len: Number of elements in the ``iterable_of_args``. When chunk_size is set to ``None`` it needs
to know the number of tasks. This can either be provided by implementing the ``__len__`` function on the
iterable object, or by specifying the number of tasks
:param max_tasks_active: Maximum number of active tasks in the queue. If ``None`` it will be converted to
``n_jobs * chunk_size * 2``
:param chunk_size: Number of simultaneous tasks to give to a worker. When ``None`` it will use ``n_splits``.
:param n_splits: Number of splits to use when ``chunk_size`` is ``None``. When both ``chunk_size`` and
``n_splits`` are ``None``, it will use ``n_splits = n_jobs * 64``.
:param worker_lifespan: Number of tasks a worker can handle before it is restarted. If ``None``, workers will
stay alive the entire time. Use this when workers use up too much memory over the course of time
:param progress_bar: When ``True`` it will display a progress bar
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param progress_bar_options: Dictionary containing keyword arguments to pass to the ``tqdm`` progress bar. See
``tqdm.tqdm()`` for details. The arguments ``total`` and ``leave`` will be overwritten by MPIRE.
:param progress_bar_style: The progress bar style to use. Can be one of ``None``, ``'std'``, or ``'notebook'``
:return: List with unordered results
"""
# Simply call imap and cast it to a list. This make sure all elements are there before returning
return list(self.imap_unordered(func, iterable_of_args, iterable_len, max_tasks_active, chunk_size,
n_splits, worker_lifespan, progress_bar, worker_init, worker_exit,
task_timeout, worker_init_timeout, worker_exit_timeout, progress_bar_options,
progress_bar_style))
[docs] def imap(self, func: Callable, iterable_of_args: Union[Sized, Iterable], iterable_len: Optional[int] = None,
max_tasks_active: Optional[int] = None, chunk_size: Optional[int] = None, n_splits: Optional[int] = None,
worker_lifespan: Optional[int] = None, progress_bar: bool = False, worker_init: Optional[Callable] = None,
worker_exit: Optional[Callable] = None, task_timeout: Optional[float] = None,
worker_init_timeout: Optional[float] = None, worker_exit_timeout: Optional[float] = None,
progress_bar_options: Optional[Dict[str, Any]] = None,
progress_bar_style: Optional[str] = None) -> Generator[Any, None, None]:
"""
Same as ``multiprocessing.imap_unordered()``, but ordered. Also allows a user to set the maximum number of
tasks available in the queue.
:param func: Function to call each time new task arguments become available. When passing on the worker ID the
function should receive the worker ID as its first argument. If shared objects are provided the function
should receive those as the next argument. If the worker state has been enabled it should receive a state
variable as the next argument
:param iterable_of_args: A numpy array or an iterable containing tuples of arguments to pass to a worker, which
passes it to the function ``func``
:param iterable_len: Number of elements in the ``iterable_of_args``. When chunk_size is set to ``None`` it needs
to know the number of tasks. This can either be provided by implementing the ``__len__`` function on the
iterable object, or by specifying the number of tasks
:param max_tasks_active: Maximum number of active tasks in the queue. If ``None`` it will be converted to
``n_jobs * chunk_size * 2``
:param chunk_size: Number of simultaneous tasks to give to a worker. When ``None`` it will use ``n_splits``.
:param n_splits: Number of splits to use when ``chunk_size`` is ``None``. When both ``chunk_size`` and
``n_splits`` are ``None``, it will use ``n_splits = n_jobs * 64``.
:param worker_lifespan: Number of tasks a worker can handle before it is restarted. If ``None``, workers will
stay alive the entire time. Use this when workers use up too much memory over the course of time
:param progress_bar: When ``True`` it will display a progress bar
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param progress_bar_options: Dictionary containing keyword arguments to pass to the ``tqdm`` progress bar. See
``tqdm.tqdm()`` for details. The arguments ``total`` and ``leave`` will be overwritten by MPIRE.
:param progress_bar_style: The progress bar style to use. Can be one of ``None``, ``'std'``, or ``'notebook'``
:return: Generator yielding ordered results
"""
# Notify workers to keep order in mind
self._worker_comms.signal_keep_order()
# If we're dealing with numpy arrays, we have to chunk them here already
if NUMPY_INSTALLED and isinstance(iterable_of_args, np.ndarray):
iterable_of_args, iterable_len, chunk_size, n_splits = apply_numpy_chunking(iterable_of_args, iterable_len,
chunk_size, n_splits,
self.pool_params.n_jobs)
# Yield results in order
next_result_idx = 0
tmp_results = {}
if iterable_len is None and hasattr(iterable_of_args, '__len__'):
iterable_len = len(iterable_of_args)
for result_idx, result in self.imap_unordered(func, ((args_idx, args) for args_idx, args
in enumerate(iterable_of_args)), iterable_len,
max_tasks_active, chunk_size, n_splits, worker_lifespan,
progress_bar, worker_init, worker_exit, task_timeout,
worker_init_timeout, worker_exit_timeout, progress_bar_options,
progress_bar_style):
# Check if the next one(s) to return is/are temporarily stored. We use a while-true block with dict.pop() to
# keep the temporary store as small as possible
while True:
if next_result_idx in tmp_results:
yield tmp_results.pop(next_result_idx)
next_result_idx += 1
else:
break
# Check if the current result is the next one to return. If so, return it
if result_idx == next_result_idx:
yield result
next_result_idx += 1
# Otherwise, temporarily store the current result
else:
tmp_results[result_idx] = result
# Yield all remaining results
for result_idx in sorted(tmp_results.keys()):
yield tmp_results.pop(result_idx)
# Notify workers to forget about order
self._worker_comms.clear_keep_order()
[docs] def imap_unordered(self, func: Callable, iterable_of_args: Union[Sized, Iterable],
iterable_len: Optional[int] = None, max_tasks_active: Optional[int] = None,
chunk_size: Optional[int] = None, n_splits: Optional[int] = None,
worker_lifespan: Optional[int] = None, progress_bar: bool = False,
worker_init: Optional[Callable] = None, worker_exit: Optional[Callable] = None,
task_timeout: Optional[float] = None, worker_init_timeout: Optional[float] = None,
worker_exit_timeout: Optional[float] = None,
progress_bar_options: Optional[Dict[str, Any]] = None,
progress_bar_style: Optional[str] = None) -> Generator[Any, None, None]:
"""
Same as ``multiprocessing.imap_unordered()``. Also allows a user to set the maximum number of tasks available in
the queue.
:param func: Function to call each time new task arguments become available. When passing on the worker ID the
function should receive the worker ID as its first argument. If shared objects are provided the function
should receive those as the next argument. If the worker state has been enabled it should receive a state
variable as the next argument
:param iterable_of_args: A numpy array or an iterable containing tuples of arguments to pass to a worker, which
passes it to the function ``func``
:param iterable_len: Number of elements in the ``iterable_of_args``. When chunk_size is set to ``None`` it needs
to know the number of tasks. This can either be provided by implementing the ``__len__`` function on the
iterable object, or by specifying the number of tasks
:param max_tasks_active: Maximum number of active tasks in the queue. If ``None`` it will be converted to
``n_jobs * chunk_size * 2``
:param chunk_size: Number of simultaneous tasks to give to a worker. When ``None`` it will use ``n_splits``.
:param n_splits: Number of splits to use when ``chunk_size`` is ``None``. When both ``chunk_size`` and
``n_splits`` are ``None``, it will use ``n_splits = n_jobs * 64``.
:param worker_lifespan: Number of tasks a worker can handle before it is restarted. If ``None``, workers will
stay alive the entire time. Use this when workers use up too much memory over the course of time
:param progress_bar: When ``True`` it will display a progress bar
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param progress_bar_options: Dictionary containing keyword arguments to pass to the ``tqdm`` progress bar. See
``tqdm.tqdm()`` for details. The arguments ``total`` and ``leave`` will be overwritten by MPIRE.
:param progress_bar_style: The progress bar style to use. Can be one of ``None``, ``'std'``, or ``'notebook'``
:return: Generator yielding unordered results
"""
# If we're dealing with numpy arrays, we have to chunk them here already
iterator_of_chunked_args = []
numpy_chunking = False
if NUMPY_INSTALLED and isinstance(iterable_of_args, np.ndarray):
iterator_of_chunked_args, iterable_len, chunk_size, n_splits = apply_numpy_chunking(
iterable_of_args, iterable_len, chunk_size, n_splits, self.pool_params.n_jobs
)
numpy_chunking = True
# Check parameters and thereby obtain the number of tasks. The chunk_size and progress bar parameters could be
# modified as well
n_tasks, max_tasks_active, chunk_size, progress_bar, progress_bar_options = check_map_parameters(
self.pool_params, iterable_of_args, iterable_len, max_tasks_active, chunk_size, n_splits, worker_lifespan,
progress_bar, progress_bar_options, progress_bar_style, task_timeout, worker_init_timeout,
worker_exit_timeout
)
new_map_params = WorkerMapParams(func, worker_init, worker_exit, worker_lifespan, progress_bar, task_timeout,
worker_init_timeout, worker_exit_timeout)
# Chunk the function arguments. Make single arguments when we're not dealing with numpy arrays
if not numpy_chunking:
iterator_of_chunked_args = chunk_tasks(iterable_of_args, n_tasks, chunk_size, n_splits)
# Grab original lock in case we have a progress bar and we need to restore it
tqdm = get_tqdm(progress_bar_style)
original_tqdm_lock = tqdm.get_lock()
tqdm_manager_owner = False
imap_iterator = None
try:
if self._map_running:
self._worker_comms.signal_exception_thrown(MAIN_PROCESS)
self._cache[MAIN_PROCESS]._set(success=False,
result=RuntimeError("Cannot call 'map' while another 'map' is running"))
self._handle_exception()
self._map_running = True
# Start tqdm manager if a progress bar is desired. Will only start one when not already started. This has to
# be done before starting the workers in case nested pools are used
if progress_bar:
tqdm_manager_owner = TqdmManager.start_manager(self.pool_params.use_dill)
# Start workers if there aren't any. If they already exist check if we need to pass on new parameters
if self._workers and not self._worker_comms.is_initialized():
logger.warning("WorkerPool parameters changed while keep_alive=True. Restarting workers.")
self.stop_and_join(keep_alive=False)
if self._workers and (self.map_params != new_map_params):
self.map_params = new_map_params
self._worker_comms.add_new_map_params(new_map_params)
if not self._workers:
self.map_params = new_map_params
self._start_workers()
# Create async result objects. The imap_iterator container will be used to store the results from the
# workers. We can yield from that
imap_iterator = UnorderedAsyncResultIterator(self._cache, n_tasks, timeout=task_timeout)
job_id = imap_iterator.job_id
# Create progress bar handler, which receives progress updates from the workers and updates the progress bar
# accordingly
with ProgressBarHandler(self.pool_params, self.map_params, progress_bar, progress_bar_options,
progress_bar_style, self._worker_comms,
self._worker_insights) as self._progress_bar_handler:
try:
# Process all args in the iterable
n_active = 0
n_tasks = 0
while True:
# Obtain next chunk of tasks
try:
chunk_of_tasks = next(iterator_of_chunked_args)
n_tasks += len(chunk_of_tasks)
except StopIteration:
break
# To keep the number of active tasks below max_tasks_active, we have to wait for results
while (not self._worker_comms.exception_thrown() and
n_active + len(chunk_of_tasks) > max_tasks_active):
try:
yield imap_iterator.next(block=True, timeout=0.01)
n_active -= 1
except queue.Empty:
pass
# If an exception has been thrown, stop now
if self._worker_comms.exception_thrown():
break
self._worker_comms.add_task(job_id, chunk_of_tasks)
n_active += len(chunk_of_tasks)
# Obtain the results not yet obtained
if not self._worker_comms.exception_thrown():
imap_iterator.set_length(n_tasks)
self._progress_bar_handler.set_new_total(n_tasks)
while not self._worker_comms.exception_thrown():
try:
yield imap_iterator.next(block=True, timeout=0.1)
except queue.Empty:
pass
except StopIteration:
break
# Terminate if exception has been thrown at this point
if self._worker_comms.exception_thrown():
self._handle_exception()
# All results are in: it's clean up time
self.stop_and_join(keep_alive=self.pool_params.keep_alive)
# Wait for the progress bar to finish, before we clean it up
if progress_bar:
self._worker_comms.wait_until_progress_bar_is_complete()
except KeyboardInterrupt:
self._handle_exception()
finally:
if tqdm_manager_owner:
tqdm.set_lock(original_tqdm_lock)
TqdmManager.stop_manager()
if imap_iterator is not None:
imap_iterator.remove_from_cache()
self._progress_bar_handler = None
self._map_running = False
self._worker_comms.reset_progress()
# Log insights
if self.pool_params.enable_insights:
logger.debug(self._worker_insights.get_insights_string())
[docs] def apply(self, func: Callable, args: Any = (), kwargs: Dict = None, callback: Optional[Callable] = None,
error_callback: Optional[Callable] = None, worker_init: Optional[Callable] = None,
worker_exit: Optional[Callable] = None, task_timeout: Optional[float] = None,
worker_init_timeout: Optional[float] = None, worker_exit_timeout: Optional[float] = None) -> Any:
"""
Apply a function to a single task. This is a blocking call.
:param func: Function to apply to the task. When passing on the worker ID the function should receive the
worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param args: Arguments to pass to a worker, which passes it to the function ``func`` as ``func(*args)``
:param kwargs: Keyword arguments to pass to a worker, which passes it to the function ``func`` as
``func(**kwargs)``
:param callback: Callback function to call when the task is finished. The callback function receives the output
of the function ``func`` as its argument
:param error_callback: Callback function to call when the task has failed. The callback function receives the
exception as its argument
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:return: Result of the function ``func`` applied to the task
"""
return self.apply_async(func, args, kwargs, callback, error_callback, worker_init, worker_exit,
task_timeout, worker_init_timeout, worker_exit_timeout).get()
[docs] def apply_async(self, func: Callable, args: Any = (), kwargs: Dict = None, callback: Optional[Callable] = None,
error_callback: Optional[Callable] = None, worker_init: Optional[Callable] = None,
worker_exit: Optional[Callable] = None, task_timeout: Optional[float] = None,
worker_init_timeout: Optional[float] = None,
worker_exit_timeout: Optional[float] = None) -> AsyncResult:
"""
Apply a function to a single task. This is a non-blocking call.
:param func: Function to apply to the task. When passing on the worker ID the function should receive the
worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param args: Arguments to pass to a worker, which passes it to the function ``func`` as ``func(*args)``
:param kwargs: Keyword arguments to pass to a worker, which passes it to the function ``func`` as
``func(**kwargs)``
:param callback: Callback function to call when the task is finished. The callback function receives the output
of the function ``func`` as its argument
:param error_callback: Callback function to call when the task has failed. The callback function receives the
exception as its argument
:param worker_init: Function to call each time a new worker starts. When passing on the worker ID the function
should receive the worker ID as its first argument. If shared objects are provided the function should
receive those as the next argument. If the worker state has been enabled it should receive a state variable
as the next argument
:param worker_exit: Function to call each time a worker exits. Return values will be fetched and made available
through :obj:`mpire.WorkerPool.get_exit_results`. When passing on the worker ID the function should receive
the worker ID as its first argument. If shared objects are provided the function should receive those as the
next argument. If the worker state has been enabled it should receive a state variable as the next argument
:param task_timeout: Timeout in seconds for a single task. When the timeout is exceeded, MPIRE will raise a
``TimeoutError``. Use ``None`` to disable (default). Note: the timeout doesn't apply to ``worker_init`` and
``worker_exit`` functions, use `worker_init_timeout` and `worker_exit_timeout` for that, respectively
:param worker_init_timeout: Timeout in seconds for the ``worker_init`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:param worker_exit_timeout: Timeout in seconds for the ``worker_exit`` function. When the timeout is exceeded,
MPIRE will raise a ``TimeoutError``. Use ``None`` to disable (default).
:return: Result of the function ``func`` applied to the task
"""
# Check if the pool has been started
if not self._workers:
self.map_params = WorkerMapParams(func, worker_init, worker_exit, None, False, task_timeout,
worker_init_timeout, worker_exit_timeout)
self._start_workers()
# Add task to the queue
result = AsyncResult(self._cache, callback, error_callback, timeout=task_timeout)
self._worker_comms.add_apply_task(result.job_id, func, args, kwargs)
return result
def _handle_exception(self) -> None:
"""
Handles exceptions thrown by workers and KeyboardInterrupts
"""
# Obtain exception
if self._worker_comms.exception_thrown():
exception = self._cache[self._worker_comms.get_exception_thrown_job_id()].get_exception()
cause = exception.__cause__
else:
self._worker_comms.signal_exception_thrown(MAIN_PROCESS)
exception = KeyboardInterrupt()
cause = exception
# Pass error to progress bar, if there is one. We are interested in the cause of the exception, as that contains
# the traceback from the worker. If there is no cause, we use the exception itself (e.g., a TimeoutError thrown
# in the timeout handler)
if self._progress_bar_handler is not None:
self._progress_bar_handler.set_exception(cause or exception)
if self._progress_bar_handler.thread is not None:
self._progress_bar_handler.thread.join()
self._progress_bar_handler = None
self.terminate()
# Clear keep order event so we can safely reuse the WorkerPool and use (i)map_unordered after an (i)map call
self._worker_comms.clear_keep_order()
# Raise
raise exception
[docs] def stop_and_join(self, keep_alive: bool = False) -> None:
"""
When ``keep_alive=False``: inserts a poison pill, grabs the exit results, waits until the tasks/results queues
are done, and waits until all workers are finished.
When ``keep_alive=True``: inserts a non-lethal poison pill, and waits until the tasks/results queues are done.
``join``and ``stop_and_join`` are aliases.
:param keep_alive: Whether to keep the workers alive
"""
if self._workers:
# All tasks have been processed and results are in. Insert (non-lethal) poison pill
if keep_alive:
self._worker_comms.insert_non_lethal_poison_pill()
else:
self._worker_comms.insert_poison_pill()
# Wait until all (non-lethal) poison pills have been consumed. When a worker's lifetime has been reached
# just before consuming the poison pill, we need to restart them
t = threading.Thread(target=self._worker_comms.join_task_queues, args=(keep_alive,))
t.daemon = True
t.start()
while not self._worker_comms.exception_thrown():
t.join(timeout=0.01)
if not t.is_alive():
break
# When an exception occurred in the above process (i.e., the worker init function raises), we need to
# handle the exception (i.e., terminate and raise)
if self._worker_comms.exception_thrown():
self._handle_exception()
# Join workers
if not keep_alive:
for wid, worker_process in enumerate(self._workers):
try:
worker_process.join()
except ValueError:
raise
# Added since Python 3.7. This will clean up any resources that are left. For some reason though,
# when using daemon processes and nested pools, a process can still be alive after the join when
# close is called and a ValueError is raised. So we wait a bit and check if the process will die. If
# not, then the GC can clean up the resources later.
if hasattr(worker_process, 'close'):
try_count = 5
while worker_process.is_alive() and try_count > 0:
time.sleep(0.01)
try_count -= 1
try:
worker_process.close()
except ValueError:
pass
self._workers = []
# Join the results queue, but do not close it. All results should be in the cache at this point (including
# exit results, because the workers joined successfully or keep_alive=True and the exit function isn't
# called), but we still need this queue for closing the results listener thread
self._worker_comms.join_results_queues(keep_alive=True)
# If an exception occurred in the exit function, we need to handle the exception (i.e., terminate and raise)
if self._worker_comms.exception_thrown():
self._handle_exception()
# Stop handler threads and join and close the results queue if we're not keeping the workers alive
if not keep_alive:
self._stop_handler_threads()
self._worker_comms.join_results_queues(keep_alive=False)
join = stop_and_join
[docs] def terminate(self) -> None:
"""
Tries to do a graceful shutdown of the workers, by interrupting them. In the case processes deadlock it will
send a sigkill.
"""
if not self._workers:
return
# Set exception thrown so workers know to stop fetching new tasks
if not self._worker_comms.exception_thrown():
self._worker_comms.signal_exception_thrown(MAIN_PROCESS)
self._cache[MAIN_PROCESS]._set(success=False, result=RuntimeError("Pool was terminated"))
# If this function is called from handle_exception, the progress bar is already terminated. If not, we need to
# terminate it here
if self._progress_bar_handler is not None:
self._progress_bar_handler.set_exception(RuntimeError("Pool was terminated"))
# When we're working with threads we have to wait for them to join. We can't kill threads in Python
if self.pool_params.start_method == 'threading':
threads = self._workers
else:
# Create cleanup threads such that processes can get killed simultaneously, which can save quite some time
threads = []
dont_wait_event = threading.Event()
dont_wait_event.set()
for worker_id in range(self.pool_params.n_jobs):
t = threading.Thread(target=self._terminate_worker, args=(worker_id, dont_wait_event))
t.start()
threads.append(t)
# Wait until cleanup threads are done
for t in threads:
t.join()
# Stop handler threads
self._stop_handler_threads()
# Drain and join the queues
self._worker_comms.drain_queues()
# Reset workers and cache. Keep only the main process, init and exit results objects
self._workers = []
self._cache = {key: self._cache[key] for key in (MAIN_PROCESS, INIT_FUNC, EXIT_FUNC)}
def _terminate_worker(self, worker_id: int, dont_wait_event: threading.Event) -> None:
"""
Terminates a single worker process.
When a process.join() raises an AssertionError, it means the worker hasn't started yet. In that case, we simply
return. A ValueError can be raised on Windows systems.
:param worker_id: Worker ID
:param dont_wait_event: Event object to indicate whether other termination threads should continue. I.e., when
we set it to False, threads should wait.
"""
# When a worker didn't start in the first place, we don't have to do anything
if self._workers[worker_id] is None or self._workers[worker_id].pid is None:
return
# Send a kill signal to the worker
self._send_kill_signal_to_worker(worker_id)
# We wait until workers are done terminating. However, we don't have all the patience in the world. When the
# patience runs out we terminate them.
try_count = 10
while try_count > 0:
try:
self._workers[worker_id].join(timeout=0.1)
if not self._workers[worker_id].is_alive():
break
except AssertionError:
return
except ValueError:
pass
# For properly joining, it can help if we try to get some results here. Workers can still be busy putting
# items in queues under the hood, even at this point.
self._worker_comms.drain_results_queue_terminate_worker(dont_wait_event)
try_count -= 1
if not dont_wait_event.is_set():
dont_wait_event.wait()
# If, after all this, the worker is still alive, we terminate it with a brutal kill signal. This shouldn't
# really happen. But, better safe than sorry
try:
if self._workers[worker_id].is_alive():
self._workers[worker_id].terminate()
self._workers[worker_id].join()
except AssertionError:
return
except ValueError:
pass
# Added since Python 3.7
if hasattr(self._workers[worker_id], 'close'):
try:
self._workers[worker_id].close()
except AssertionError:
return
except ValueError:
pass
def _send_kill_signal_to_worker(self, worker_id: int) -> None:
"""
Sends a kill signal to a worker process, but only if we know it's running a task.
:param worker_id: Worker ID
"""
# Signal handling in Windows is cumbersome, to say the least. Therefore, it handles error handling
# differently. See Worker::_exit_gracefully_windows for more information.
if not RUNNING_WINDOWS and self.pool_params.start_method != "threading":
with self._worker_comms.get_worker_running_task_lock(worker_id):
if self._worker_comms.get_worker_running_task(worker_id):
# A signal should only be send once
self._worker_comms.set_worker_running_task(worker_id, False)
# Send signal
try:
os.kill(self._workers[worker_id].pid, signal.SIGUSR1)
except (ProcessLookupError, ValueError):
pass
def _stop_handler_threads(self) -> None:
"""
Stops results, restart, timeout, and unexpected death handler threads.
"""
self._handler_threads_stop_event.set()
# Join results listener thread
if self._results_handler_thread is not None and self._results_handler_thread.is_alive():
self._worker_comms.insert_poison_pill_results_listener()
self._results_handler_thread.join()
self._results_handler_thread = None
# Join restart thread
if self._restart_handler_thread is not None and self._restart_handler_thread.is_alive():
self._worker_comms.signal_worker_restart_condition()
self._restart_handler_thread.join()
self._restart_handler_thread = None
# Join timeout handler thread
if self._timeout_handler_thread is not None and self._timeout_handler_thread.is_alive():
self._timeout_handler_thread.join()
self._timeout_handler_thread = None
# Join unexpected death handler thread
if (self._unexpected_death_handler_thread is not None and
self._unexpected_death_handler_thread.is_alive()):
self._unexpected_death_handler_thread.join()
self._unexpected_death_handler_thread = None
[docs] def print_insights(self) -> None:
"""
Prints insights per worker
"""
print(self._worker_insights.get_insights_string())
[docs] def get_insights(self) -> Dict:
"""
Creates insights from the raw insight data
:return: Dictionary containing worker insights
"""
return self._worker_insights.get_insights()