Convert master_manager_launcher to use multiprocessing.

infra/libs/process_invocation/multiprocess.py

+# Copyright 2015 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""Convenience functions for code calling the multiprocessing module."""
+
+import contextlib
+import multiprocessing
+
+
+@contextlib.contextmanager
+def MultiPool(processes):
+  """Manages a multiprocessing.Pool making sure to close the pool when done.
+
+  This will also call pool.terminate() when an exception is raised (and
+  re-raised the exception to the calling procedure can handle it).
+
+  If you plan on using a multiprocess pool, take a look at
+  http://bugs.python.org/issue12157 and
+  http://stackoverflow.com/a/1408476/3984761.
+  """
+  try:
+    pool = multiprocessing.Pool(processes=processes)
+    yield pool
+    pool.close()
+  except:
+    pool.terminate()
+    raise
+  finally:
+    pool.join()
+
+
+def safe_map(func, args, processes):  # pragma: no cover
+  """Executes a function over multiple sets of arguments in parallel.
+
+  This works around two gotchas easily encountered when using
+  multiprocessing.Pool.map_async(). Specifically, that map_async() can hang if
+  the arguent count is an empty list, and that KeyboardInterrupts are not
+  handled properly.
+
+  func must be a real, top-level function (cannot be a nested function or
+       lambda, see http://stackoverflow.com/q/8804830/3984761). It should take
+       a single argument.
+
+  args is an iterable of arguments over which repeated func calls are made.
+
+  processes is the number of processes to use for execution.
+  """
+
+  # Prevent map from hanging, see http://bugs.python.org/issue12157.
+  if not args:
+    return []
+
+  with MultiPool(processes) as pool:
+    # This strange invocation is so ctrl-c can interrupt the map_async. See
+    # http://stackoverflow.com/a/1408476/3984761 for details.
+    return pool.map_async(func, args).get(9999999)