nonblocking-itertools 0.1.2

Five functions in itertools can be substantially slowed when their input iterables generate data with lag. This might occur when the generator yields data being streamed from the internet, disk, or a multithreaded or multiprocessing workflow.

The most substantial slowdowns are in permutations, combinations, combinations_with_replacement, and product. These block the flow of execution until their input iterable(s) have completely finished being converted to a tuple. Example:

import time
import itertools
import nonblocking_itertools

def laggy_generator(iterable):
    for i in iterable:
        time.sleep(1)
        yield i

print("Testing itertools.combinations")
start = time.time()
data = laggy_generator(range(4))
for combo in itertools.combinations(data, 2):
    print("Seconds from start:", int(time.time() - start), "Combination:", combo)

print("Testing nonblocking_itertools.combinations")
start = time.time()
data = laggy_generator(range(4))
for combo in nonblocking_itertools.combinations(data, 2):
    print("Seconds from start:", time.time() - start, "Combination:", combo)

Output:

Testing itertools.combinations
Seconds from start: 4 Combination: (0, 1)
Seconds from start: 4 Combination: (0, 2)
Seconds from start: 4 Combination: (0, 3)
Seconds from start: 4 Combination: (1, 2)
Seconds from start: 4 Combination: (1, 3)
Seconds from start: 4 Combination: (2, 3)
Testing nonblocking_itertools.combinations
Seconds from start: 2 Combination: (0, 1)
Seconds from start: 3 Combination: (0, 2)
Seconds from start: 3 Combination: (1, 2)
Seconds from start: 4 Combination: (0, 3)
Seconds from start: 4 Combination: (1, 3)
Seconds from start: 4 Combination: (2, 3)

Notice that all results from itertools.combinations were printed after 4 seconds, while the results from nonblocking_itertools.combinations were printed as soon as it was possible to yield them, after 2 or 3 seconds. Note that nonblocking_itertools does not generally return results in the same order as the equivalent itertools methods.

The other improvement in nonblocking_itertools is with the chain method, for when iterables have some parallelizeable processing that is initiated only during iteration. The itertools.chain method only starts iterating through later iterables when the first is exhausted. That means it misses the opportunity to kick off parallel processing in the later iterables. By contrast, nonblocking_itertools initiates immediate and simultaneous multithreaded iteration through iterables to take advantage of opportunities for parallelization. Its output is identical to itertools.chain.

Example:

import time
import itertools
import nonblocking_itertools

def laggy_generator(iterable):
    for i in iterable:
        time.sleep(1)
        yield i

print("Testing itertools.chain")
start = time.time()
data1 = laggy_generator(range(2))
data2 = laggy_generator(range(2, 4))
for element in itertools.chain(data1, data2):
    print("Seconds from start:", int(time.time() - start), "Element:", element)

print("Testing nonblocking_itertools.chain")
start = time.time()
data1 = laggy_generator(range(2))
data2 = laggy_generator(range(2, 4))
for element in nonblocking_itertools.chain(data1, data2):
    print("Seconds from start:", int(time.time() - start), "Element:", element)

Output:

Testing itertools.chain
Seconds from start: 1 Element: 0
Seconds from start: 2 Element: 1
Seconds from start: 3 Element: 2
Seconds from start: 4 Element: 3
Testing nonblocking_itertools.chain
Seconds from start: 1 Element: 0
Seconds from start: 2 Element: 1
Seconds from start: 2 Element: 2
Seconds from start: 2 Element: 3

Notice that nonblocking_itertools.chain finishes after 2 seconds, while itertools.chain takes 4 seconds to finish.