You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 

56 lines
1.9 KiB

"""Utility functions.
"""
import contextlib
import multiprocessing
from milc import cli
@contextlib.contextmanager
def parallelize():
"""Returns a function that can be used in place of a map() call.
Attempts to use `mpire`, falling back to `multiprocessing` if it's not
available. If parallelization is not requested, returns the original map()
function.
"""
# Work out if we've already got a config value for parallel searching
if cli.config.user.parallel_search is None:
parallel_search = True
else:
parallel_search = cli.config.user.parallel_search
# Non-parallel searches use `map()`
if not parallel_search:
yield map
return
# Prefer mpire's `WorkerPool` if it's available
with contextlib.suppress(ImportError):
from mpire import WorkerPool
from mpire.utils import make_single_arguments
with WorkerPool() as pool:
def _worker(func, *args):
# Ensure we don't unpack tuples -- mpire's `WorkerPool` tries to do so normally so we tell it not to.
for r in pool.imap_unordered(func, make_single_arguments(*args, generator=False), progress_bar=True):
yield r
yield _worker
return
# Otherwise fall back to multiprocessing's `Pool`
with multiprocessing.Pool() as pool:
yield pool.imap_unordered
def parallel_map(*args, **kwargs):
"""Effectively runs `map()` but executes it in parallel if necessary.
"""
with parallelize() as map_fn:
# This needs to be enclosed in a `list()` as some implementations return
# a generator function, which means the scope of the pool is closed off
# before the results are returned. Returning a list ensures results are
# materialised before any worker pool is shut down.
return list(map_fn(*args, **kwargs))