"""
Helpers for embarrassingly parallel code.
"""
# Author: Gael Varoquaux < gael dot varoquaux at normalesup dot org >
# Copyright: 2010, Gael Varoquaux
# License: BSD 3 clause
from __future__ import division
import os
import sys
from math import sqrt
import functools
import collections
import time
import threading
import itertools
from uuid import uuid4
from numbers import Integral
import warnings
import queue
import weakref
from contextlib import nullcontext
from multiprocessing import TimeoutError
from ._multiprocessing_helpers import mp
from .logger import Logger, short_format_time
from .disk import memstr_to_bytes
from ._parallel_backends import (FallbackToBackend, MultiprocessingBackend,
ThreadingBackend, SequentialBackend,
LokyBackend)
from ._utils import eval_expr, _Sentinel
# Make sure that those two classes are part of the public joblib.parallel API
# so that 3rd party backend implementers can import them from here.
from ._parallel_backends import AutoBatchingMixin # noqa
from ._parallel_backends import ParallelBackendBase # noqa
IS_PYPY = hasattr(sys, "pypy_version_info")
BACKENDS = {
'threading': ThreadingBackend,
'sequential': SequentialBackend,
}
# name of the backend used by default by Parallel outside of any context
# managed by ``parallel_config`` or ``parallel_backend``.
# threading is the only backend that is always everywhere
DEFAULT_BACKEND = 'threading'
MAYBE_AVAILABLE_BACKENDS = {'multiprocessing', 'loky'}
# if multiprocessing is available, so is loky, we set it as the default
# backend
if mp is not None:
BACKENDS['multiprocessing'] = MultiprocessingBackend
from .externals import loky
BACKENDS['loky'] = LokyBackend
DEFAULT_BACKEND = 'loky'
DEFAULT_THREAD_BACKEND = 'threading'
# Thread local value that can be overridden by the ``parallel_config`` context
# manager
_backend = threading.local()
def _register_dask():
"""Register Dask Backend if called with parallel_config(backend="dask")"""
try:
from ._dask import DaskDistributedBackend
register_parallel_backend('dask', DaskDistributedBackend)
except ImportError as e:
msg = ("To use the dask.distributed backend you must install both "
"the `dask` and distributed modules.\n\n"
"See https://dask.pydata.org/en/latest/install.html for more "
"information.")
raise ImportError(msg) from e
EXTERNAL_BACKENDS = {
'dask': _register_dask,
}
# Sentinels for the default values of the Parallel constructor and
# the parallel_config and parallel_backend context managers
default_parallel_config = {
"backend": _Sentinel(default_value=None),
"n_jobs": _Sentinel(default_value=None),
"verbose": _Sentinel(default_value=0),
"temp_folder": _Sentinel(default_value=None),
"max_nbytes": _Sentinel(default_value="1M"),
"mmap_mode": _Sentinel(default_value="r"),
"prefer": _Sentinel(default_value=None),
"require": _Sentinel(default_value=None),
}
VALID_BACKEND_HINTS = ('processes', 'threads', None)
VALID_BACKEND_CONSTRAINTS = ('sharedmem', None)
def _get_config_param(param, context_config, key):
"""Return the value of a parallel config parameter
Explicitly setting it in Parallel has priority over setting in a
parallel_(config/backend) context manager.
"""
if param is not default_parallel_config[key]:
# param is explicitly set, return it
return param
if context_config[key] is not default_parallel_config[key]:
# there's a context manager and the key is set, return it
return context_config[key]
# Otherwise, we are in the default_parallel_config,
# return the default value
return param.default_value
def get_active_backend(
prefer=default_parallel_config["prefer"],
require=default_parallel_config["require"],
verbose=default_parallel_config["verbose"],
):
"""Return the active default backend"""
backend, config = _get_active_backend(prefer, require, verbose)
n_jobs = _get_config_param(
default_parallel_config['n_jobs'], config, "n_jobs"
)
return backend, n_jobs
def _get_active_backend(
prefer=default_parallel_config["prefer"],
require=default_parallel_config["require"],
verbose=default_parallel_config["verbose"],
):
"""Return the active default backend"""
backend_config = getattr(_backend, "config", default_parallel_config)
backend = _get_config_param(
default_parallel_config['backend'], backend_config, "backend"
)
prefer = _get_config_param(prefer, backend_config, "prefer")
require = _get_config_param(require, backend_config, "require")
verbose = _get_config_param(verbose, backend_config, "verbose")
if prefer not in VALID_BACKEND_HINTS:
raise ValueError(
f"prefer={prefer} is not a valid backend hint, "
f"expected one of {VALID_BACKEND_HINTS}"
)
if require not in VALID_BACKEND_CONSTRAINTS:
raise ValueError(
f"require={require} is not a valid backend constraint, "
f"expected one of {VALID_BACKEND_CONSTRAINTS}"
)
if prefer == 'processes' and require == 'sharedmem':
raise ValueError(
"prefer == 'processes' and require == 'sharedmem'"
" are inconsistent settings"
)
explicit_backend = True
if backend is None:
# We are either outside of the scope of any parallel_(config/backend)
# context manager or the context manager did not set a backend.
# create the default backend instance now.
backend = BACKENDS[DEFAULT_BACKEND](nesting_level=0)
explicit_backend = False
# Try to use the backend set by the user with the context manager.
nesting_level = backend.nesting_level
uses_threads = getattr(backend, 'uses_threads', False)
supports_sharedmem = getattr(backend, 'supports_sharedmem', False)
# Force to use thread-based backend if the provided backend does not
# match the shared memory constraint or if the backend is not explicitly
# given and threads are preferred.
force_threads = (require == 'sharedmem' and not supports_sharedmem)
force_threads |= (
not explicit_backend and prefer == 'threads' and not uses_threads
)
if force_threads:
# This backend does not match the shared memory constraint:
# fallback to the default thead-based backend.
sharedmem_backend = BACKENDS[DEFAULT_THREAD_BACKEND](
nesting_level=nesting_level
)
# Warn the user if we forced the backend to thread-based, while the
# user explicitly specified a non-thread-based backend.
if verbose >= 10 and explicit_backend:
print(
f"Using {sharedmem_backend.__class__.__name__} as "
f"joblib backend instead of {backend.__class__.__name__} "
"as the latter does not provide shared memory semantics."
)
# Force to n_jobs=1 by default
thread_config = backend_config.copy()
thread_config['n_jobs'] = 1
return sharedmem_backend, thread_config
return backend, backend_config
class parallel_config:
"""Set the default backend or configuration for :class:`~joblib.Parallel`.
This is an alternative to directly passing keyword arguments to the
:class:`~joblib.Parallel` class constructor. It is particularly useful when
calling into library code that uses joblib internally but does not expose
the various parallel configuration arguments in its own API.
Parameters
----------
backend: str or ParallelBackendBase instance, default=None
If ``backend`` is a string it must match a previously registered
implementation using the :func:`~register_parallel_backend` function.
By default the following backends are available:
- 'loky': single-host, process-based parallelism (used by default),
- 'threading': single-host, thread-based parallelism,
- 'multiprocessing': legacy single-host, process-based parallelism.
'loky' is recommended to run functions that manipulate Python objects.
'threading' is a low-overhead alternative that is most efficient for
functions that release the Global Interpreter Lock: e.g. I/O-bound
code or CPU-bound code in a few calls to native code that explicitly
releases the GIL. Note that on some rare systems (such as pyodide),
multiprocessing and loky may not be available, in which case joblib
defaults to threading.
In addition, if the ``dask`` and ``distributed`` Python packages are
installed, it is possible to use the 'dask' backend for better
scheduling of nested parallel calls without over-subscription and
potentially distribute parallel calls over a networked cluster of
several hosts.
It is also possible to use the distributed 'ray' backend for
distributing the workload to a cluster of nodes. See more details
in the Examples section below.
Alternatively the backend can be passed directly as an instance.
n_jobs: int, default=None
The maximum number of concurrently running jobs, such as the number
of Python worker processes when ``backend="loky"`` or the size of the
thread-pool when ``backend="threading"``.
This argument is converted to an integer, rounded below for float.
If -1 is given, `joblib` tries to use all CPUs. The number of CPUs
``n_cpus`` is obtained with :func:`~cpu_count`.
For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. For instance,
using ``n_jobs=-2`` will result in all CPUs but one being used.
This argument can also go above ``n_cpus``, which will cause
oversubscription. In some cases, slight oversubscription can be
beneficial, e.g., for tasks with large I/O operations.
If 1 is given, no parallel computing code is used at all, and the
behavior amounts to a simple python `for` loop. This mode is not
compatible with `timeout`.
None is a marker for 'unset' that will be interpreted as n_jobs=1
unless the call is performed under a :func:`~parallel_config`
context manager that sets another value for ``n_jobs``.
If n_jobs = 0 then a ValueError is raised.
verbose: int, default=0
The verbosity level: if non zero, progress messages are
printed. Above 50, the output is sent to stdout.
The frequency of the messages increases with the verbosity level.
If it more than 10, all iterations are reported.
temp_folder: str or None, default=None
Folder to be used by the pool for memmapping large arrays
for sharing memory with worker processes. If None, this will try in
order:
- a folder pointed by the ``JOBLIB_TEMP_FOLDER`` environment
variable,
- ``/dev/shm`` if the folder exists and is writable: this is a
RAM disk filesystem available by default on modern Linux
distributions,
- the default system temporary folder that can be
overridden with ``TMP``, ``TMPDIR`` or ``TEMP`` environment
variables, typically ``/tmp`` under Unix operating systems.
max_nbytes int, str, or None, optional, default='1M'
Threshold on the size of arrays passed to the workers that
triggers automated memory mapping in temp_folder. Can be an int
in Bytes, or a human-readable string, e.g., '1M' for 1 megabyte.
Use None to disable memmapping of large arrays.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}, default='r'
Memmapping mode for numpy arrays passed to workers. None will
disable memmapping, other modes defined in the numpy.memmap doc:
https://numpy.org/doc/stable/reference/generated/numpy.memmap.html
Also, see 'max_nbytes' parameter documentation for more details.
prefer: str in {'processes', 'threads'} or None, default=None
Soft hint to choose the default backend.
The default process-based backend is 'loky' and the default
thread-based backend is 'threading'. Ignored if the ``backend``
parameter is specified.
require: 'sharedmem' or None, default=None
Hard constraint to select the backend. If set to 'sharedmem',
the selected backend will be single-host and thread-based.
inner_max_num_threads: int, default=None
If not None, overwrites the limit set on the number of threads
usable in some third-party library threadpools like OpenBLAS,
MKL or OpenMP. This is only used with the ``loky`` backend.
backend_params: dict
Additional parameters to pass to the backend constructor when
backend is a string.
Notes
-----
Joblib tries to limit the oversubscription by limiting the number of
threads usable in some third-party library threadpools like OpenBLAS, MKL
or OpenMP. The default limit in each worker is set to
``max(cpu_count() // effective_n_jobs, 1)`` but this limit can be
overwritten with the ``inner_max_num_threads`` argument which will be used
to set this limit in the child processes.
.. versionadded:: 1.3
Examples
--------
>>> from operator import neg
>>> with parallel_config(backend='threading'):
... print(Parallel()(delayed(neg)(i + 1) for i in range(5)))
...
[-1, -2, -3, -4, -5]
To use the 'ray' joblib backend add the following lines:
>>> from ray.util.joblib import register_ray # doctest: +SKIP
>>> register_ray() # doctest: +SKIP
>>> with parallel_config(backend="ray"): # doctest: +SKIP
... print(Parallel()(delayed(neg)(i + 1) for i in range(5)))
[-1, -2, -3, -4, -5]
"""
def __init__(
self,
backend=default_parallel_config["backend"],
*,
n_jobs=default_parallel_config["n_jobs"],
verbose=default_parallel_config["verbose"],
temp_folder=default_parallel_config["temp_folder"],
max_nbytes=default_parallel_config["max_nbytes"],
mmap_mode=default_parallel_config["mmap_mode"],
prefer=default_parallel_config["prefer"],
require=default_parallel_config["require"],
inner_max_num_threads=None,
**backend_params
):
# Save the parallel info and set the active parallel config
self.old_parallel_config = getattr(
_backend, "config", default_parallel_config
)
backend = self._check_backend(
backend, inner_max_num_threads, **backend_params
)
new_config = {
"n_jobs": n_jobs,
"verbose": verbose,
"temp_folder": temp_folder,
"max_nbytes": max_nbytes,
"mmap_mode": mmap_mode,
"prefer": prefer,
"require": require,
"backend": backend
}
self.parallel_config = self.old_parallel_config.copy()
self.parallel_config.update({
k: v for k, v in new_config.items()
if not isinstance(v, _Sentinel)
})
setattr(_backend, "config", self.parallel_config)
def _check_backend(self, backend, inner_max_num_threads, **backend_params):
if backend is default_parallel_config['backend']:
if inner_max_num_threads is not None or len(backend_params) > 0:
raise ValueError(
"inner_max_num_threads and other constructor "
"parameters backend_params are only supported "
"when backend is not None."
)
return backend
if isinstance(backend, str):
# Handle non-registered or missing backends
if backend not in BACKENDS:
if backend in EXTERNAL_BACKENDS:
register = EXTERNAL_BACKENDS[backend]
register()
elif backend in MAYBE_AVAILABLE_BACKENDS:
warnings.warn(
f"joblib backend '{backend}' is not available on "
f"your system, falling back to {DEFAULT_BACKEND}.",
UserWarning,
stacklevel=2
)
BACKENDS[backend] = BACKENDS[DEFAULT_BACKEND]
else:
raise ValueError(
f"Invalid backend: {backend}, expected one of "
f"{sorted(BACKENDS.keys())}"
)
backend = BACKENDS[backend](**backend_params)
if inner_max_num_threads is not None:
msg = (
f"{backend.__class__.__name__} does not accept setting the "
"inner_max_num_threads argument."
)
assert backend.supports_inner_max_num_threads, msg
backend.inner_max_num_threads = inner_max_num_threads
# If the nesting_level of the backend is not set previously, use the
# nesting level from the previous active_backend to set it
if backend.nesting_level is None:
parent_backend = self.old_parallel_config['backend']
if parent_backend is default_parallel_config['backend']:
nesting_level = 0
else:
nesting_level = parent_backend.nesting_level
backend.nesting_level = nesting_level
return backend
def __enter__(self):
return self.parallel_config
def __exit__(self, type, value, traceback):
self.unregister()
def unregister(self):
setattr(_backend, "config", self.old_parallel_config)
class parallel_backend(parallel_config):
"""Change the default backend used by Parallel inside a with block.
.. warning::
It is advised to use the :class:`~joblib.parallel_config` context
manager instead, which allows more fine-grained control over the
backend configuration.
If ``backend`` is a string it must match a previously registered
implementation using the :func:`~register_parallel_backend` function.
By default the following backends are available:
- 'loky': single-host, process-based parallelism (used by default),
- 'threading': single-host, thread-based parallelism,
- 'multiprocessing': legacy single-host, process-based parallelism.
'loky' is recommended to run functions that manipulate Python objects.
'threading' is a low-overhead alternative that is most efficient for
functions that release the Global Interpreter Lock: e.g. I/O-bound code or
CPU-bound code in a few calls to native code that explicitly releases the
GIL. Note that on some rare systems (such as Pyodide),
multiprocessing and loky may not be available, in which case joblib
defaults to threading.
You can also use the `Dask `_ joblib
backend to distribute work across machines. This works well with
scikit-learn estimators with the ``n_jobs`` parameter, for example::
>>> import joblib # doctest: +SKIP
>>> from sklearn.model_selection import GridSearchCV # doctest: +SKIP
>>> from dask.distributed import Client, LocalCluster # doctest: +SKIP
>>> # create a local Dask cluster
>>> cluster = LocalCluster() # doctest: +SKIP
>>> client = Client(cluster) # doctest: +SKIP
>>> grid_search = GridSearchCV(estimator, param_grid, n_jobs=-1)
... # doctest: +SKIP
>>> with joblib.parallel_backend("dask", scatter=[X, y]): # doctest: +SKIP
... grid_search.fit(X, y)
It is also possible to use the distributed 'ray' backend for distributing
the workload to a cluster of nodes. To use the 'ray' joblib backend add
the following lines::
>>> from ray.util.joblib import register_ray # doctest: +SKIP
>>> register_ray() # doctest: +SKIP
>>> with parallel_backend("ray"): # doctest: +SKIP
... print(Parallel()(delayed(neg)(i + 1) for i in range(5)))
[-1, -2, -3, -4, -5]
Alternatively the backend can be passed directly as an instance.
By default all available workers will be used (``n_jobs=-1``) unless the
caller passes an explicit value for the ``n_jobs`` parameter.
This is an alternative to passing a ``backend='backend_name'`` argument to
the :class:`~Parallel` class constructor. It is particularly useful when
calling into library code that uses joblib internally but does not expose
the backend argument in its own API.
>>> from operator import neg
>>> with parallel_backend('threading'):
... print(Parallel()(delayed(neg)(i + 1) for i in range(5)))
...
[-1, -2, -3, -4, -5]
Joblib also tries to limit the oversubscription by limiting the number of
threads usable in some third-party library threadpools like OpenBLAS, MKL
or OpenMP. The default limit in each worker is set to
``max(cpu_count() // effective_n_jobs, 1)`` but this limit can be
overwritten with the ``inner_max_num_threads`` argument which will be used
to set this limit in the child processes.
.. versionadded:: 0.10
See Also
--------
joblib.parallel_config: context manager to change the backend
configuration.
"""
def __init__(self, backend, n_jobs=-1, inner_max_num_threads=None,
**backend_params):
super().__init__(
backend=backend,
n_jobs=n_jobs,
inner_max_num_threads=inner_max_num_threads,
**backend_params
)
if self.old_parallel_config is None:
self.old_backend_and_jobs = None
else:
self.old_backend_and_jobs = (
self.old_parallel_config["backend"],
self.old_parallel_config["n_jobs"],
)
self.new_backend_and_jobs = (
self.parallel_config["backend"],
self.parallel_config["n_jobs"],
)
def __enter__(self):
return self.new_backend_and_jobs
# Under Linux or OS X the default start method of multiprocessing
# can cause third party libraries to crash. Under Python 3.4+ it is possible
# to set an environment variable to switch the default start method from
# 'fork' to 'forkserver' or 'spawn' to avoid this issue albeit at the cost
# of causing semantic changes and some additional pool instantiation overhead.
DEFAULT_MP_CONTEXT = None
if hasattr(mp, 'get_context'):
method = os.environ.get('JOBLIB_START_METHOD', '').strip() or None
if method is not None:
DEFAULT_MP_CONTEXT = mp.get_context(method=method)
class BatchedCalls(object):
"""Wrap a sequence of (func, args, kwargs) tuples as a single callable"""
def __init__(self, iterator_slice, backend_and_jobs, reducer_callback=None,
pickle_cache=None):
self.items = list(iterator_slice)
self._size = len(self.items)
self._reducer_callback = reducer_callback
if isinstance(backend_and_jobs, tuple):
self._backend, self._n_jobs = backend_and_jobs
else:
# this is for backward compatibility purposes. Before 0.12.6,
# nested backends were returned without n_jobs indications.
self._backend, self._n_jobs = backend_and_jobs, None
self._pickle_cache = pickle_cache if pickle_cache is not None else {}
def __call__(self):
# Set the default nested backend to self._backend but do not set the
# change the default number of processes to -1
with parallel_config(backend=self._backend, n_jobs=self._n_jobs):
return [func(*args, **kwargs)
for func, args, kwargs in self.items]
def __reduce__(self):
if self._reducer_callback is not None:
self._reducer_callback()
# no need to pickle the callback.
return (
BatchedCalls,
(self.items, (self._backend, self._n_jobs), None,
self._pickle_cache)
)
def __len__(self):
return self._size
# Possible exit status for a task
TASK_DONE = "Done"
TASK_ERROR = "Error"
TASK_PENDING = "Pending"
###############################################################################
# CPU count that works also when multiprocessing has been disabled via
# the JOBLIB_MULTIPROCESSING environment variable
def cpu_count(only_physical_cores=False):
"""Return the number of CPUs.
This delegates to loky.cpu_count that takes into account additional
constraints such as Linux CFS scheduler quotas (typically set by container
runtimes such as docker) and CPU affinity (for instance using the taskset
command on Linux).
If only_physical_cores is True, do not take hyperthreading / SMT logical
cores into account.
"""
if mp is None:
return 1
return loky.cpu_count(only_physical_cores=only_physical_cores)
###############################################################################
# For verbosity
def _verbosity_filter(index, verbose):
""" Returns False for indices increasingly apart, the distance
depending on the value of verbose.
We use a lag increasing as the square of index
"""
if not verbose:
return True
elif verbose > 10:
return False
if index == 0:
return False
verbose = .5 * (11 - verbose) ** 2
scale = sqrt(index / verbose)
next_scale = sqrt((index + 1) / verbose)
return (int(next_scale) == int(scale))
###############################################################################
def delayed(function):
"""Decorator used to capture the arguments of a function."""
def delayed_function(*args, **kwargs):
return function, args, kwargs
try:
delayed_function = functools.wraps(function)(delayed_function)
except AttributeError:
" functools.wraps fails on some callable objects "
return delayed_function
###############################################################################
class BatchCompletionCallBack(object):
"""Callback to keep track of completed results and schedule the next tasks.
This callable is executed by the parent process whenever a worker process
has completed a batch of tasks.
It is used for progress reporting, to update estimate of the batch
processing duration and to schedule the next batch of tasks to be
processed.
It is assumed that this callback will always be triggered by the backend
right after the end of a task, in case of success as well as in case of
failure.
"""
##########################################################################
# METHODS CALLED BY THE MAIN THREAD #
##########################################################################
def __init__(self, dispatch_timestamp, batch_size, parallel):
self.dispatch_timestamp = dispatch_timestamp
self.batch_size = batch_size
self.parallel = parallel
self.parallel_call_id = parallel._call_id
# Internals to keep track of the status and outcome of the task.
# Used to hold a reference to the future-like object returned by the
# backend after launching this task
# This will be set later when calling `register_job`, as it is only
# created once the task has been submitted.
self.job = None
if not parallel._backend.supports_retrieve_callback:
# The status is only used for asynchronous result retrieval in the
# callback.
self.status = None
else:
# The initial status for the job is TASK_PENDING.
# Once it is done, it will be either TASK_DONE, or TASK_ERROR.
self.status = TASK_PENDING
def register_job(self, job):
"""Register the object returned by `apply_async`."""
self.job = job
def get_result(self, timeout):
"""Returns the raw result of the task that was submitted.
If the task raised an exception rather than returning, this same
exception will be raised instead.
If the backend supports the retrieval callback, it is assumed that this
method is only called after the result has been registered. It is
ensured by checking that `self.status(timeout)` does not return
TASK_PENDING. In this case, `get_result` directly returns the
registered result (or raise the registered exception).
For other backends, there are no such assumptions, but `get_result`
still needs to synchronously retrieve the result before it can
return it or raise. It will block at most `self.timeout` seconds
waiting for retrieval to complete, after that it raises a TimeoutError.
"""
backend = self.parallel._backend
if backend.supports_retrieve_callback:
# We assume that the result has already been retrieved by the
# callback thread, and is stored internally. It's just waiting to
# be returned.
return self._return_or_raise()
# For other backends, the main thread needs to run the retrieval step.
try:
if backend.supports_timeout:
result = self.job.get(timeout=timeout)
else:
result = self.job.get()
outcome = dict(result=result, status=TASK_DONE)
except BaseException as e:
outcome = dict(result=e, status=TASK_ERROR)
self._register_outcome(outcome)
return self._return_or_raise()
def _return_or_raise(self):
try:
if self.status == TASK_ERROR:
raise self._result
return self._result
finally:
del self._result
def get_status(self, timeout):
"""Get the status of the task.
This function also checks if the timeout has been reached and register
the TimeoutError outcome when it is the case.
"""
if timeout is None or self.status != TASK_PENDING:
return self.status
# The computation are running and the status is pending.
# Check that we did not wait for this jobs more than `timeout`.
now = time.time()
if not hasattr(self, "_completion_timeout_counter"):
self._completion_timeout_counter = now
if (now - self._completion_timeout_counter) > timeout:
outcome = dict(result=TimeoutError(), status=TASK_ERROR)
self._register_outcome(outcome)
return self.status
##########################################################################
# METHODS CALLED BY CALLBACK THREADS #
##########################################################################
def __call__(self, out):
"""Function called by the callback thread after a job is completed."""
# If the backend doesn't support callback retrievals, the next batch of
# tasks is dispatched regardless. The result will be retrieved by the
# main thread when calling `get_result`.
if not self.parallel._backend.supports_retrieve_callback:
self._dispatch_new()
return
# If the backend supports retrieving the result in the callback, it
# registers the task outcome (TASK_ERROR or TASK_DONE), and schedules
# the next batch if needed.
with self.parallel._lock:
# Edge case where while the task was processing, the `parallel`
# instance has been reset and a new call has been issued, but the
# worker managed to complete the task and trigger this callback
# call just before being aborted by the reset.
if self.parallel._call_id != self.parallel_call_id:
return
# When aborting, stop as fast as possible and do not retrieve the
# result as it won't be returned by the Parallel call.
if self.parallel._aborting:
return
# Retrieves the result of the task in the main process and dispatch
# a new batch if needed.
job_succeeded = self._retrieve_result(out)
if not self.parallel.return_ordered:
# Append the job to the queue in the order of completion
# instead of submission.
self.parallel._jobs.append(self)
if job_succeeded:
self._dispatch_new()
def _dispatch_new(self):
"""Schedule the next batch of tasks to be processed."""
# This steps ensure that auto-batching works as expected.
this_batch_duration = time.time() - self.dispatch_timestamp
self.parallel._backend.batch_completed(self.batch_size,
this_batch_duration)
# Schedule the next batch of tasks.
with self.parallel._lock:
self.parallel.n_completed_tasks += self.batch_size
self.parallel.print_progress()
if self.parallel._original_iterator is not None:
self.parallel.dispatch_next()
def _retrieve_result(self, out):
"""Fetch and register the outcome of a task.
Return True if the task succeeded, False otherwise.
This function is only called by backends that support retrieving
the task result in the callback thread.
"""
try:
result = self.parallel._backend.retrieve_result_callback(out)
outcome = dict(status=TASK_DONE, result=result)
except BaseException as e:
# Avoid keeping references to parallel in the error.
e.__traceback__ = None
outcome = dict(result=e, status=TASK_ERROR)
self._register_outcome(outcome)
return outcome['status'] != TASK_ERROR
##########################################################################
# This method can be called either in the main thread #
# or in the callback thread. #
##########################################################################
def _register_outcome(self, outcome):
"""Register the outcome of a task.
This method can be called only once, future calls will be ignored.
"""
# Covers the edge case where the main thread tries to register a
# `TimeoutError` while the callback thread tries to register a result
# at the same time.
with self.parallel._lock:
if self.status not in (TASK_PENDING, None):
return
self.status = outcome["status"]
self._result = outcome["result"]
# Once the result and the status are extracted, the last reference to
# the job can be deleted.
self.job = None
# As soon as an error as been spotted, early stopping flags are sent to
# the `parallel` instance.
if self.status == TASK_ERROR:
self.parallel._exception = True
self.parallel._aborting = True
###############################################################################
def register_parallel_backend(name, factory, make_default=False):
"""Register a new Parallel backend factory.
The new backend can then be selected by passing its name as the backend
argument to the :class:`~Parallel` class. Moreover, the default backend can
be overwritten globally by setting make_default=True.
The factory can be any callable that takes no argument and return an
instance of ``ParallelBackendBase``.
Warning: this function is experimental and subject to change in a future
version of joblib.
.. versionadded:: 0.10
"""
BACKENDS[name] = factory
if make_default:
global DEFAULT_BACKEND
DEFAULT_BACKEND = name
def effective_n_jobs(n_jobs=-1):
"""Determine the number of jobs that can actually run in parallel
n_jobs is the number of workers requested by the callers. Passing n_jobs=-1
means requesting all available workers for instance matching the number of
CPU cores on the worker host(s).
This method should return a guesstimate of the number of workers that can
actually perform work concurrently with the currently enabled default
backend. The primary use case is to make it possible for the caller to know
in how many chunks to slice the work.
In general working on larger data chunks is more efficient (less scheduling
overhead and better use of CPU cache prefetching heuristics) as long as all
the workers have enough work to do.
Warning: this function is experimental and subject to change in a future
version of joblib.
.. versionadded:: 0.10
"""
if n_jobs == 1:
return 1
backend, backend_n_jobs = get_active_backend()
if n_jobs is None:
n_jobs = backend_n_jobs
return backend.effective_n_jobs(n_jobs=n_jobs)
###############################################################################
class Parallel(Logger):
''' Helper class for readable parallel mapping.
Read more in the :ref:`User Guide `.
Parameters
----------
n_jobs: int, default=None
The maximum number of concurrently running jobs, such as the number
of Python worker processes when ``backend="loky"`` or the size of
the thread-pool when ``backend="threading"``.
This argument is converted to an integer, rounded below for float.
If -1 is given, `joblib` tries to use all CPUs. The number of CPUs
``n_cpus`` is obtained with :func:`~cpu_count`.
For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. For instance,
using ``n_jobs=-2`` will result in all CPUs but one being used.
This argument can also go above ``n_cpus``, which will cause
oversubscription. In some cases, slight oversubscription can be
beneficial, e.g., for tasks with large I/O operations.
If 1 is given, no parallel computing code is used at all, and the
behavior amounts to a simple python `for` loop. This mode is not
compatible with ``timeout``.
None is a marker for 'unset' that will be interpreted as n_jobs=1
unless the call is performed under a :func:`~parallel_config`
context manager that sets another value for ``n_jobs``.
If n_jobs = 0 then a ValueError is raised.
backend: str, ParallelBackendBase instance or None, default='loky'
Specify the parallelization backend implementation.
Supported backends are:
- "loky" used by default, can induce some
communication and memory overhead when exchanging input and
output data with the worker Python processes. On some rare
systems (such as Pyiodide), the loky backend may not be
available.
- "multiprocessing" previous process-based backend based on
`multiprocessing.Pool`. Less robust than `loky`.
- "threading" is a very low-overhead backend but it suffers
from the Python Global Interpreter Lock if the called function
relies a lot on Python objects. "threading" is mostly useful
when the execution bottleneck is a compiled extension that
explicitly releases the GIL (for instance a Cython loop wrapped
in a "with nogil" block or an expensive call to a library such
as NumPy).
- finally, you can register backends by calling
:func:`~register_parallel_backend`. This will allow you to
implement a backend of your liking.
It is not recommended to hard-code the backend name in a call to
:class:`~Parallel` in a library. Instead it is recommended to set
soft hints (prefer) or hard constraints (require) so as to make it
possible for library users to change the backend from the outside
using the :func:`~parallel_config` context manager.
return_as: str in {'list', 'generator', 'generator_unordered'}, default='list'
If 'list', calls to this instance will return a list, only when
all results have been processed and retrieved.
If 'generator', it will return a generator that yields the results
as soon as they are available, in the order the tasks have been
submitted with.
If 'generator_unordered', the generator will immediately yield
available results independently of the submission order. The output
order is not deterministic in this case because it depends on the
concurrency of the workers.
prefer: str in {'processes', 'threads'} or None, default=None
Soft hint to choose the default backend if no specific backend
was selected with the :func:`~parallel_config` context manager.
The default process-based backend is 'loky' and the default
thread-based backend is 'threading'. Ignored if the ``backend``
parameter is specified.
require: 'sharedmem' or None, default=None
Hard constraint to select the backend. If set to 'sharedmem',
the selected backend will be single-host and thread-based even
if the user asked for a non-thread based backend with
:func:`~joblib.parallel_config`.
verbose: int, default=0
The verbosity level: if non zero, progress messages are
printed. Above 50, the output is sent to stdout.
The frequency of the messages increases with the verbosity level.
If it more than 10, all iterations are reported.
timeout: float or None, default=None
Timeout limit for each task to complete. If any task takes longer
a TimeOutError will be raised. Only applied when n_jobs != 1
pre_dispatch: {'all', integer, or expression, as in '3*n_jobs'}, default='2*n_jobs'
The number of batches (of tasks) to be pre-dispatched.
Default is '2*n_jobs'. When batch_size="auto" this is reasonable
default and the workers should never starve. Note that only basic
arithmetics are allowed here and no modules can be used in this
expression.
batch_size: int or 'auto', default='auto'
The number of atomic tasks to dispatch at once to each
worker. When individual evaluations are very fast, dispatching
calls to workers can be slower than sequential computation because
of the overhead. Batching fast computations together can mitigate
this.
The ``'auto'`` strategy keeps track of the time it takes for a
batch to complete, and dynamically adjusts the batch size to keep
the time on the order of half a second, using a heuristic. The
initial batch size is 1.
``batch_size="auto"`` with ``backend="threading"`` will dispatch
batches of a single task at a time as the threading backend has
very little overhead and using larger batch size has not proved to
bring any gain in that case.
temp_folder: str or None, default=None
Folder to be used by the pool for memmapping large arrays
for sharing memory with worker processes. If None, this will try in
order:
- a folder pointed by the JOBLIB_TEMP_FOLDER environment
variable,
- /dev/shm if the folder exists and is writable: this is a
RAM disk filesystem available by default on modern Linux
distributions,
- the default system temporary folder that can be
overridden with TMP, TMPDIR or TEMP environment
variables, typically /tmp under Unix operating systems.
Only active when ``backend="loky"`` or ``"multiprocessing"``.
max_nbytes int, str, or None, optional, default='1M'
Threshold on the size of arrays passed to the workers that
triggers automated memory mapping in temp_folder. Can be an int
in Bytes, or a human-readable string, e.g., '1M' for 1 megabyte.
Use None to disable memmapping of large arrays.
Only active when ``backend="loky"`` or ``"multiprocessing"``.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}, default='r'
Memmapping mode for numpy arrays passed to workers. None will
disable memmapping, other modes defined in the numpy.memmap doc:
https://numpy.org/doc/stable/reference/generated/numpy.memmap.html
Also, see 'max_nbytes' parameter documentation for more details.
Notes
-----
This object uses workers to compute in parallel the application of a
function to many different arguments. The main functionality it brings
in addition to using the raw multiprocessing or concurrent.futures API
are (see examples for details):
* More readable code, in particular since it avoids
constructing list of arguments.
* Easier debugging:
- informative tracebacks even when the error happens on
the client side
- using 'n_jobs=1' enables to turn off parallel computing
for debugging without changing the codepath
- early capture of pickling errors
* An optional progress meter.
* Interruption of multiprocesses jobs with 'Ctrl-C'
* Flexible pickling control for the communication to and from
the worker processes.
* Ability to use shared memory efficiently with worker
processes for large numpy-based datastructures.
Note that the intended usage is to run one call at a time. Multiple
calls to the same Parallel object will result in a ``RuntimeError``
Examples
--------
A simple example:
>>> from math import sqrt
>>> from joblib import Parallel, delayed
>>> Parallel(n_jobs=1)(delayed(sqrt)(i**2) for i in range(10))
[0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0]
Reshaping the output when the function has several return
values:
>>> from math import modf
>>> from joblib import Parallel, delayed
>>> r = Parallel(n_jobs=1)(delayed(modf)(i/2.) for i in range(10))
>>> res, i = zip(*r)
>>> res
(0.0, 0.5, 0.0, 0.5, 0.0, 0.5, 0.0, 0.5, 0.0, 0.5)
>>> i
(0.0, 0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 3.0, 4.0, 4.0)
The progress meter: the higher the value of `verbose`, the more
messages:
>>> from time import sleep
>>> from joblib import Parallel, delayed
>>> r = Parallel(n_jobs=2, verbose=10)(
... delayed(sleep)(.2) for _ in range(10)) #doctest: +SKIP
[Parallel(n_jobs=2)]: Done 1 tasks | elapsed: 0.6s
[Parallel(n_jobs=2)]: Done 4 tasks | elapsed: 0.8s
[Parallel(n_jobs=2)]: Done 10 out of 10 | elapsed: 1.4s finished
Traceback example, note how the line of the error is indicated
as well as the values of the parameter passed to the function that
triggered the exception, even though the traceback happens in the
child process:
>>> from heapq import nlargest
>>> from joblib import Parallel, delayed
>>> Parallel(n_jobs=2)(
... delayed(nlargest)(2, n) for n in (range(4), 'abcde', 3))
... # doctest: +SKIP
-----------------------------------------------------------------------
Sub-process traceback:
-----------------------------------------------------------------------
TypeError Mon Nov 12 11:37:46 2012
PID: 12934 Python 2.7.3: /usr/bin/python
........................................................................
/usr/lib/python2.7/heapq.pyc in nlargest(n=2, iterable=3, key=None)
419 if n >= size:
420 return sorted(iterable, key=key, reverse=True)[:n]
421
422 # When key is none, use simpler decoration
423 if key is None:
--> 424 it = izip(iterable, count(0,-1)) # decorate
425 result = _nlargest(n, it)
426 return map(itemgetter(0), result) # undecorate
427
428 # General case, slowest method
TypeError: izip argument #1 must support iteration
_______________________________________________________________________
Using pre_dispatch in a producer/consumer situation, where the
data is generated on the fly. Note how the producer is first
called 3 times before the parallel loop is initiated, and then
called to generate new data on the fly:
>>> from math import sqrt
>>> from joblib import Parallel, delayed
>>> def producer():
... for i in range(6):
... print('Produced %s' % i)
... yield i
>>> out = Parallel(n_jobs=2, verbose=100, pre_dispatch='1.5*n_jobs')(
... delayed(sqrt)(i) for i in producer()) #doctest: +SKIP
Produced 0
Produced 1
Produced 2
[Parallel(n_jobs=2)]: Done 1 jobs | elapsed: 0.0s
Produced 3
[Parallel(n_jobs=2)]: Done 2 jobs | elapsed: 0.0s
Produced 4
[Parallel(n_jobs=2)]: Done 3 jobs | elapsed: 0.0s
Produced 5
[Parallel(n_jobs=2)]: Done 4 jobs | elapsed: 0.0s
[Parallel(n_jobs=2)]: Done 6 out of 6 | elapsed: 0.0s remaining: 0.0s
[Parallel(n_jobs=2)]: Done 6 out of 6 | elapsed: 0.0s finished
''' # noqa: E501
def __init__(
self,
n_jobs=default_parallel_config["n_jobs"],
backend=default_parallel_config['backend'],
return_as="list",
verbose=default_parallel_config["verbose"],
timeout=None,
pre_dispatch='2 * n_jobs',
batch_size='auto',
temp_folder=default_parallel_config["temp_folder"],
max_nbytes=default_parallel_config["max_nbytes"],
mmap_mode=default_parallel_config["mmap_mode"],
prefer=default_parallel_config["prefer"],
require=default_parallel_config["require"],
):
# Initiate parent Logger class state
super().__init__()
# Interpret n_jobs=None as 'unset'
if n_jobs is None:
n_jobs = default_parallel_config["n_jobs"]
active_backend, context_config = _get_active_backend(
prefer=prefer, require=require, verbose=verbose
)
nesting_level = active_backend.nesting_level
self.verbose = _get_config_param(verbose, context_config, "verbose")
self.timeout = timeout
self.pre_dispatch = pre_dispatch
if return_as not in {"list", "generator", "generator_unordered"}:
raise ValueError(
'Expected `return_as` parameter to be a string equal to "list"'
f',"generator" or "generator_unordered", but got {return_as} '
"instead."
)
self.return_as = return_as
self.return_generator = return_as != "list"
self.return_ordered = return_as != "generator_unordered"
# Check if we are under a parallel_config or parallel_backend
# context manager and use the config from the context manager
# for arguments that are not explicitly set.
self._backend_args = {
k: _get_config_param(param, context_config, k) for param, k in [
(max_nbytes, "max_nbytes"),
(temp_folder, "temp_folder"),
(mmap_mode, "mmap_mode"),
(prefer, "prefer"),
(require, "require"),
(verbose, "verbose"),
]
}
if isinstance(self._backend_args["max_nbytes"], str):
self._backend_args["max_nbytes"] = memstr_to_bytes(
self._backend_args["max_nbytes"]
)
self._backend_args["verbose"] = max(
0, self._backend_args["verbose"] - 50
)
if DEFAULT_MP_CONTEXT is not None:
self._backend_args['context'] = DEFAULT_MP_CONTEXT
elif hasattr(mp, "get_context"):
self._backend_args['context'] = mp.get_context()
if backend is default_parallel_config['backend'] or backend is None:
backend = active_backend
elif isinstance(backend, ParallelBackendBase):
# Use provided backend as is, with the current nesting_level if it
# is not set yet.
if backend.nesting_level is None:
backend.nesting_level = nesting_level
elif hasattr(backend, 'Pool') and hasattr(backend, 'Lock'):
# Make it possible to pass a custom multiprocessing context as
# backend to change the start method to forkserver or spawn or
# preload modules on the forkserver helper process.
self._backend_args['context'] = backend
backend = MultiprocessingBackend(nesting_level=nesting_level)
elif backend not in BACKENDS and backend in MAYBE_AVAILABLE_BACKENDS:
warnings.warn(
f"joblib backend '{backend}' is not available on "
f"your system, falling back to {DEFAULT_BACKEND}.",
UserWarning,
stacklevel=2)
BACKENDS[backend] = BACKENDS[DEFAULT_BACKEND]
backend = BACKENDS[DEFAULT_BACKEND](nesting_level=nesting_level)
else:
try:
backend_factory = BACKENDS[backend]
except KeyError as e:
raise ValueError("Invalid backend: %s, expected one of %r"
% (backend, sorted(BACKENDS.keys()))) from e
backend = backend_factory(nesting_level=nesting_level)
n_jobs = _get_config_param(n_jobs, context_config, "n_jobs")
if n_jobs is None:
# No specific context override and no specific value request:
# default to the default of the backend.
n_jobs = backend.default_n_jobs
try:
n_jobs = int(n_jobs)
except ValueError:
raise ValueError("n_jobs could not be converted to int")
self.n_jobs = n_jobs
if (require == 'sharedmem' and
not getattr(backend, 'supports_sharedmem', False)):
raise ValueError("Backend %s does not support shared memory"
% backend)
if (batch_size == 'auto' or isinstance(batch_size, Integral) and
batch_size > 0):
self.batch_size = batch_size
else:
raise ValueError(
"batch_size must be 'auto' or a positive integer, got: %r"
% batch_size)
if not isinstance(backend, SequentialBackend):
if self.return_generator and not backend.supports_return_generator:
raise ValueError(
"Backend {} does not support "
"return_as={}".format(backend, return_as)
)
# This lock is used to coordinate the main thread of this process
# with the async callback thread of our the pool.
self._lock = threading.RLock()
self._jobs = collections.deque()
self._pending_outputs = list()
self._ready_batches = queue.Queue()
self._reducer_callback = None
# Internal variables
self._backend = backend
self._running = False
self._managed_backend = False
self._id = uuid4().hex
self._call_ref = None
def __enter__(self):
self._managed_backend = True
self._calling = False
self._initialize_backend()
return self
def __exit__(self, exc_type, exc_value, traceback):
self._managed_backend = False
if self.return_generator and self._calling:
self._abort()
self._terminate_and_reset()
def _initialize_backend(self):
"""Build a process or thread pool and return the number of workers"""
try:
n_jobs = self._backend.configure(n_jobs=self.n_jobs, parallel=self,
**self._backend_args)
if self.timeout is not None and not self._backend.supports_timeout:
warnings.warn(
'The backend class {!r} does not support timeout. '
"You have set 'timeout={}' in Parallel but "
"the 'timeout' parameter will not be used.".format(
self._backend.__class__.__name__,
self.timeout))
except FallbackToBackend as e:
# Recursively initialize the backend in case of requested fallback.
self._backend = e.backend
n_jobs = self._initialize_backend()
return n_jobs
def _effective_n_jobs(self):
if self._backend:
return self._backend.effective_n_jobs(self.n_jobs)
return 1
def _terminate_and_reset(self):
if hasattr(self._backend, 'stop_call') and self._calling:
self._backend.stop_call()
self._calling = False
if not self._managed_backend:
self._backend.terminate()
def _dispatch(self, batch):
"""Queue the batch for computing, with or without multiprocessing
WARNING: this method is not thread-safe: it should be only called
indirectly via dispatch_one_batch.
"""
# If job.get() catches an exception, it closes the queue:
if self._aborting:
return
batch_size = len(batch)
self.n_dispatched_tasks += batch_size
self.n_dispatched_batches += 1
dispatch_timestamp = time.time()
batch_tracker = BatchCompletionCallBack(
dispatch_timestamp, batch_size, self
)
if self.return_ordered:
self._jobs.append(batch_tracker)
# If return_ordered is False, the batch_tracker is not stored in the
# jobs queue at the time of submission. Instead, it will be appended to
# the queue by itself as soon as the callback is triggered to be able
# to return the results in the order of completion.
job = self._backend.apply_async(batch, callback=batch_tracker)
batch_tracker.register_job(job)
def dispatch_next(self):
"""Dispatch more data for parallel processing
This method is meant to be called concurrently by the multiprocessing
callback. We rely on the thread-safety of dispatch_one_batch to protect
against concurrent consumption of the unprotected iterator.
"""
if not self.dispatch_one_batch(self._original_iterator):
self._iterating = False
self._original_iterator = None
def dispatch_one_batch(self, iterator):
"""Prefetch the tasks for the next batch and dispatch them.
The effective size of the batch is computed here.
If there are no more jobs to dispatch, return False, else return True.
The iterator consumption and dispatching is protected by the same
lock so calling this function should be thread safe.
"""
if self._aborting:
return False
batch_size = self._get_batch_size()
with self._lock:
# to ensure an even distribution of the workload between workers,
# we look ahead in the original iterators more than batch_size
# tasks - However, we keep consuming only one batch at each
# dispatch_one_batch call. The extra tasks are stored in a local
# queue, _ready_batches, that is looked-up prior to re-consuming
# tasks from the origal iterator.
try:
tasks = self._ready_batches.get(block=False)
except queue.Empty:
# slice the iterator n_jobs * batchsize items at a time. If the
# slice returns less than that, then the current batchsize puts
# too much weight on a subset of workers, while other may end
# up starving. So in this case, re-scale the batch size
# accordingly to distribute evenly the last items between all
# workers.
n_jobs = self._cached_effective_n_jobs
big_batch_size = batch_size * n_jobs
try:
islice = list(itertools.islice(iterator, big_batch_size))
except Exception as e:
# Handle the fact that the generator of task raised an
# exception. As this part of the code can be executed in
# a thread internal to the backend, register a task with
# an error that will be raised in the user's thread.
if isinstance(e.__context__, queue.Empty):
# Suppress the cause of the exception if it is
# queue.Empty to avoid cluttered traceback. Only do it
# if the __context__ is really empty to avoid messing
# with causes of the original error.
e.__cause__ = None
batch_tracker = BatchCompletionCallBack(
0, batch_size, self
)
self._jobs.append(batch_tracker)
batch_tracker._register_outcome(dict(
result=e, status=TASK_ERROR
))
return True
if len(islice) == 0:
return False
elif (iterator is self._original_iterator and
len(islice) < big_batch_size):
# We reached the end of the original iterator (unless
# iterator is the ``pre_dispatch``-long initial slice of
# the original iterator) -- decrease the batch size to
# account for potential variance in the batches running
# time.
final_batch_size = max(1, len(islice) // (10 * n_jobs))
else:
final_batch_size = max(1, len(islice) // n_jobs)
# enqueue n_jobs batches in a local queue
for i in range(0, len(islice), final_batch_size):
tasks = BatchedCalls(islice[i:i + final_batch_size],
self._backend.get_nested_backend(),
self._reducer_callback,
self._pickle_cache)
self._ready_batches.put(tasks)
# finally, get one task.
tasks = self._ready_batches.get(block=False)
if len(tasks) == 0:
# No more tasks available in the iterator: tell caller to stop.
return False
else:
self._dispatch(tasks)
return True
def _get_batch_size(self):
"""Returns the effective batch size for dispatch"""
if self.batch_size == 'auto':
return self._backend.compute_batch_size()
else:
# Fixed batch size strategy
return self.batch_size
def _print(self, msg):
"""Display the message on stout or stderr depending on verbosity"""
# XXX: Not using the logger framework: need to
# learn to use logger better.
if not self.verbose:
return
if self.verbose < 50:
writer = sys.stderr.write
else:
writer = sys.stdout.write
writer(f"[{self}]: {msg}\n")
def _is_completed(self):
"""Check if all tasks have been completed"""
return self.n_completed_tasks == self.n_dispatched_tasks and not (
self._iterating or self._aborting
)
def print_progress(self):
"""Display the process of the parallel execution only a fraction
of time, controlled by self.verbose.
"""
if not self.verbose:
return
elapsed_time = time.time() - self._start_time
if self._is_completed():
# Make sure that we get a last message telling us we are done
self._print(
f"Done {self.n_completed_tasks:3d} out of "
f"{self.n_completed_tasks:3d} | elapsed: "
f"{short_format_time(elapsed_time)} finished"
)
return
# Original job iterator becomes None once it has been fully
# consumed: at this point we know the total number of jobs and we are
# able to display an estimation of the remaining time based on already
# completed jobs. Otherwise, we simply display the number of completed
# tasks.
elif self._original_iterator is not None:
if _verbosity_filter(self.n_dispatched_batches, self.verbose):
return
self._print(
f"Done {self.n_completed_tasks:3d} tasks | elapsed: "
f"{short_format_time(elapsed_time)}"
)
else:
index = self.n_completed_tasks
# We are finished dispatching
total_tasks = self.n_dispatched_tasks
# We always display the first loop
if not index == 0:
# Display depending on the number of remaining items
# A message as soon as we finish dispatching, cursor is 0
cursor = (total_tasks - index + 1 -
self._pre_dispatch_amount)
frequency = (total_tasks // self.verbose) + 1
is_last_item = (index + 1 == total_tasks)
if (is_last_item or cursor % frequency):
return
remaining_time = (elapsed_time / index) * \
(self.n_dispatched_tasks - index * 1.0)
# only display status if remaining time is greater or equal to 0
self._print(
f"Done {index:3d} out of {total_tasks:3d} | elapsed: "
f"{short_format_time(elapsed_time)} remaining: "
f"{short_format_time(remaining_time)}"
)
def _abort(self):
# Stop dispatching new jobs in the async callback thread
self._aborting = True
# If the backend allows it, cancel or kill remaining running
# tasks without waiting for the results as we will raise
# the exception we got back to the caller instead of returning
# any result.
backend = self._backend
if (not self._aborted and hasattr(backend, 'abort_everything')):
# If the backend is managed externally we need to make sure
# to leave it in a working state to allow for future jobs
# scheduling.
ensure_ready = self._managed_backend
backend.abort_everything(ensure_ready=ensure_ready)
self._aborted = True
def _start(self, iterator, pre_dispatch):
# Only set self._iterating to True if at least a batch
# was dispatched. In particular this covers the edge
# case of Parallel used with an exhausted iterator. If
# self._original_iterator is None, then this means either
# that pre_dispatch == "all", n_jobs == 1 or that the first batch
# was very quick and its callback already dispatched all the
# remaining jobs.
self._iterating = False
if self.dispatch_one_batch(iterator):
self._iterating = self._original_iterator is not None
while self.dispatch_one_batch(iterator):
pass
if pre_dispatch == "all":
# The iterable was consumed all at once by the above for loop.
# No need to wait for async callbacks to trigger to
# consumption.
self._iterating = False
def _get_outputs(self, iterator, pre_dispatch):
"""Iterator returning the tasks' output as soon as they are ready."""
dispatch_thread_id = threading.get_ident()
detach_generator_exit = False
try:
self._start(iterator, pre_dispatch)
# first yield returns None, for internal use only. This ensures
# that we enter the try/except block and start dispatching the
# tasks.
yield
with self._backend.retrieval_context():
yield from self._retrieve()
except GeneratorExit:
# The generator has been garbage collected before being fully
# consumed. This aborts the remaining tasks if possible and warn
# the user if necessary.
self._exception = True
# In some interpreters such as PyPy, GeneratorExit can be raised in
# a different thread than the one used to start the dispatch of the
# parallel tasks. This can lead to hang when a thread attempts to
# join itself. As workaround, we detach the execution of the
# aborting code to a dedicated thread. We then need to make sure
# the rest of the function does not call `_terminate_and_reset`
# in finally.
if dispatch_thread_id != threading.get_ident():
if not IS_PYPY:
warnings.warn(
"A generator produced by joblib.Parallel has been "
"gc'ed in an unexpected thread. This behavior should "
"not cause major -issues but to make sure, please "
"report this warning and your use case at "
"https://github.com/joblib/joblib/issues so it can "
"be investigated."
)
detach_generator_exit = True
_parallel = self
class _GeneratorExitThread(threading.Thread):
def run(self):
_parallel._abort()
if _parallel.return_generator:
_parallel._warn_exit_early()
_parallel._terminate_and_reset()
_GeneratorExitThread(
name="GeneratorExitThread"
).start()
return
# Otherwise, we are in the thread that started the dispatch: we can
# safely abort the execution and warn the user.
self._abort()
if self.return_generator:
self._warn_exit_early()
raise
# Note: we catch any BaseException instead of just Exception instances
# to also include KeyboardInterrupt
except BaseException:
self._exception = True
self._abort()
raise
finally:
# Store the unconsumed tasks and terminate the workers if necessary
_remaining_outputs = ([] if self._exception else self._jobs)
self._jobs = collections.deque()
self._running = False
if not detach_generator_exit:
self._terminate_and_reset()
while len(_remaining_outputs) > 0:
batched_results = _remaining_outputs.popleft()
batched_results = batched_results.get_result(self.timeout)
for result in batched_results:
yield result
def _wait_retrieval(self):
"""Return True if we need to continue retrieving some tasks."""
# If the input load is still being iterated over, it means that tasks
# are still on the dispatch waitlist and their results will need to
# be retrieved later on.
if self._iterating:
return True
# If some of the dispatched tasks are still being processed by the
# workers, wait for the compute to finish before starting retrieval
if self.n_completed_tasks < self.n_dispatched_tasks:
return True
# For backends that does not support retrieving asynchronously the
# result to the main process, all results must be carefully retrieved
# in the _retrieve loop in the main thread while the backend is alive.
# For other backends, the actual retrieval is done asynchronously in
# the callback thread, and we can terminate the backend before the
# `self._jobs` result list has been emptied. The remaining results
# will be collected in the `finally` step of the generator.
if not self._backend.supports_retrieve_callback:
if len(self._jobs) > 0:
return True
return False
def _retrieve(self):
while self._wait_retrieval():
# If the callback thread of a worker has signaled that its task
# triggered an exception, or if the retrieval loop has raised an
# exception (e.g. `GeneratorExit`), exit the loop and surface the
# worker traceback.
if self._aborting:
self._raise_error_fast()
break
# If the next job is not ready for retrieval yet, we just wait for
# async callbacks to progress.
if ((len(self._jobs) == 0) or
(self._jobs[0].get_status(
timeout=self.timeout) == TASK_PENDING)):
time.sleep(0.01)
continue
# We need to be careful: the job list can be filling up as
# we empty it and Python list are not thread-safe by
# default hence the use of the lock
with self._lock:
batched_results = self._jobs.popleft()
# Flatten the batched results to output one output at a time
batched_results = batched_results.get_result(self.timeout)
for result in batched_results:
self._nb_consumed += 1
yield result
def _raise_error_fast(self):
"""If we are aborting, raise if a job caused an error."""
# Find the first job whose status is TASK_ERROR if it exists.
with self._lock:
error_job = next((job for job in self._jobs
if job.status == TASK_ERROR), None)
# If this error job exists, immediately raise the error by
# calling get_result. This job might not exists if abort has been
# called directly or if the generator is gc'ed.
if error_job is not None:
error_job.get_result(self.timeout)
def _warn_exit_early(self):
"""Warn the user if the generator is gc'ed before being consumned."""
ready_outputs = self.n_completed_tasks - self._nb_consumed
is_completed = self._is_completed()
msg = ""
if ready_outputs:
msg += (
f"{ready_outputs} tasks have been successfully executed "
" but not used."
)
if not is_completed:
msg += " Additionally, "
if not is_completed:
msg += (
f"{self.n_dispatched_tasks - self.n_completed_tasks} tasks "
"which were still being processed by the workers have been "
"cancelled."
)
if msg:
msg += (
" You could benefit from adjusting the input task "
"iterator to limit unnecessary computation time."
)
warnings.warn(msg)
def _get_sequential_output(self, iterable):
"""Separate loop for sequential output.
This simplifies the traceback in case of errors and reduces the
overhead of calling sequential tasks with `joblib`.
"""
try:
self._iterating = True
self._original_iterator = iterable
batch_size = self._get_batch_size()
if batch_size != 1:
it = iter(iterable)
iterable_batched = iter(
lambda: tuple(itertools.islice(it, batch_size)), ()
)
iterable = (
task for batch in iterable_batched for task in batch
)
# first yield returns None, for internal use only. This ensures
# that we enter the try/except block and setup the generator.
yield None
# Sequentially call the tasks and yield the results.
for func, args, kwargs in iterable:
self.n_dispatched_batches += 1
self.n_dispatched_tasks += 1
res = func(*args, **kwargs)
self.n_completed_tasks += 1
self.print_progress()
yield res
self._nb_consumed += 1
except BaseException:
self._exception = True
self._aborting = True
self._aborted = True
raise
finally:
self.print_progress()
self._running = False
self._iterating = False
self._original_iterator = None
def _reset_run_tracking(self):
"""Reset the counters and flags used to track the execution."""
# Makes sur the parallel instance was not previously running in a
# thread-safe way.
with getattr(self, '_lock', nullcontext()):
if self._running:
msg = 'This Parallel instance is already running !'
if self.return_generator is True:
msg += (
" Before submitting new tasks, you must wait for the "
"completion of all the previous tasks, or clean all "
"references to the output generator."
)
raise RuntimeError(msg)
self._running = True
# Counter to keep track of the task dispatched and completed.
self.n_dispatched_batches = 0
self.n_dispatched_tasks = 0
self.n_completed_tasks = 0
# Following count is incremented by one each time the user iterates
# on the output generator, it is used to prepare an informative
# warning message in case the generator is deleted before all the
# dispatched tasks have been consumed.
self._nb_consumed = 0
# Following flags are used to synchronize the threads in case one of
# the tasks error-out to ensure that all workers abort fast and that
# the backend terminates properly.
# Set to True as soon as a worker signals that a task errors-out
self._exception = False
# Set to True in case of early termination following an incident
self._aborting = False
# Set to True after abortion is complete
self._aborted = False
def __call__(self, iterable):
"""Main function to dispatch parallel tasks."""
self._reset_run_tracking()
self._start_time = time.time()
if not self._managed_backend:
n_jobs = self._initialize_backend()
else:
n_jobs = self._effective_n_jobs()
if n_jobs == 1:
# If n_jobs==1, run the computation sequentially and return
# immediately to avoid overheads.
output = self._get_sequential_output(iterable)
next(output)
return output if self.return_generator else list(output)
# Let's create an ID that uniquely identifies the current call. If the
# call is interrupted early and that the same instance is immediately
# re-used, this id will be used to prevent workers that were
# concurrently finalizing a task from the previous call to run the
# callback.
with self._lock:
self._call_id = uuid4().hex
# self._effective_n_jobs should be called in the Parallel.__call__
# thread only -- store its value in an attribute for further queries.
self._cached_effective_n_jobs = n_jobs
if isinstance(self._backend, LokyBackend):
# For the loky backend, we add a callback executed when reducing
# BatchCalls, that makes the loky executor use a temporary folder
# specific to this Parallel object when pickling temporary memmaps.
# This callback is necessary to ensure that several Parallel
# objects using the same reusable executor don't use the same
# temporary resources.
def _batched_calls_reducer_callback():
# Relevant implementation detail: the following lines, called
# when reducing BatchedCalls, are called in a thread-safe
# situation, meaning that the context of the temporary folder
# manager will not be changed in between the callback execution
# and the end of the BatchedCalls pickling. The reason is that
# pickling (the only place where set_current_context is used)
# is done from a single thread (the queue_feeder_thread).
self._backend._workers._temp_folder_manager.set_current_context( # noqa
self._id
)
self._reducer_callback = _batched_calls_reducer_callback
# self._effective_n_jobs should be called in the Parallel.__call__
# thread only -- store its value in an attribute for further queries.
self._cached_effective_n_jobs = n_jobs
backend_name = self._backend.__class__.__name__
if n_jobs == 0:
raise RuntimeError("%s has no active worker." % backend_name)
self._print(
f"Using backend {backend_name} with {n_jobs} concurrent workers."
)
if hasattr(self._backend, 'start_call'):
self._backend.start_call()
# Following flag prevents double calls to `backend.stop_call`.
self._calling = True
iterator = iter(iterable)
pre_dispatch = self.pre_dispatch
if pre_dispatch == 'all':
# prevent further dispatch via multiprocessing callback thread
self._original_iterator = None
self._pre_dispatch_amount = 0
else:
self._original_iterator = iterator
if hasattr(pre_dispatch, 'endswith'):
pre_dispatch = eval_expr(
pre_dispatch.replace("n_jobs", str(n_jobs))
)
self._pre_dispatch_amount = pre_dispatch = int(pre_dispatch)
# The main thread will consume the first pre_dispatch items and
# the remaining items will later be lazily dispatched by async
# callbacks upon task completions.
# TODO: this iterator should be batch_size * n_jobs
iterator = itertools.islice(iterator, self._pre_dispatch_amount)
# Use a caching dict for callables that are pickled with cloudpickle to
# improve performances. This cache is used only in the case of
# functions that are defined in the __main__ module, functions that
# are defined locally (inside another function) and lambda expressions.
self._pickle_cache = dict()
output = self._get_outputs(iterator, pre_dispatch)
self._call_ref = weakref.ref(output)
# The first item from the output is blank, but it makes the interpreter
# progress until it enters the Try/Except block of the generator and
# reaches the first `yield` statement. This starts the asynchronous
# dispatch of the tasks to the workers.
next(output)
return output if self.return_generator else list(output)
def __repr__(self):
return '%s(n_jobs=%s)' % (self.__class__.__name__, self.n_jobs)