import concurrent.futures from typing import Any, Callable, List, TypeVar T = TypeVar('T') class Parallel: """ Utilities for executing parallel operations. This is used as a convenience so that we don't have to plumb async/await support (yuck) through the network, but we can still make multiple queries at once to remote services and the DB. """ @staticmethod def execute(lambdas: List[Callable[[], Any]]) -> List[Any]: """ Given a list of callables, execute them and return a list of their returns. Guarantees order of return based on order of callable. """ if len(lambdas) == 0: return [] with concurrent.futures.ThreadPoolExecutor(max_workers=len(lambdas)) as executor: futures = {executor.submit(lambdas[pos]): pos for pos in range(len(lambdas))} results = [] # List: Tuple[Any, int] for future in concurrent.futures.as_completed(futures): pos = futures[future] data = future.result() results.append((data, pos)) return [r[0] for r in sorted(results, key=lambda r: r[1])] @staticmethod def map(lam: Callable[[T], Any], params: List[T]) -> List[Any]: """ Given a callable and a list of params, executes that callable with each set of params in the list and returns a list of their returns. Guarantees order of return. """ if len(params) == 0: return [] with concurrent.futures.ThreadPoolExecutor(max_workers=len(params)) as executor: futures = {executor.submit(lam, params[pos]): pos for pos in range(len(params))} results = [] # List: Tuple[Any, int] for future in concurrent.futures.as_completed(futures): pos = futures[future] data = future.result() results.append((data, pos)) return [r[0] for r in sorted(results, key=lambda r: r[1])] @staticmethod def call(lambdas: 'List[Callable[..., Any]]', *params: Any) -> List[Any]: """ Given a list of callables and zero or more params, calls each callable in parallel with the params specified. Essentially a map of params to multiple callables in parallel. Returns a list of returns, garanteed to be in the same order as the lambdas. """ if len(lambdas) == 0: return [] with concurrent.futures.ThreadPoolExecutor(max_workers=len(lambdas)) as executor: futures = {executor.submit(lambdas[pos], *params): pos for pos in range(len(lambdas))} results = [] # List: Tuple[Any, int] for future in concurrent.futures.as_completed(futures): pos = futures[future] data = future.result() results.append((data, pos)) return [r[0] for r in sorted(results, key=lambda r: r[1])] @staticmethod def flatten(lists: List[List[Any]]) -> List[Any]: """ Convenience function that probably exists in functools, but whatever. Takes a list of lists, and returns a list made of all those lists joined together. """ return [item for sublist in lists for item in sublist]