3 from __future__ import annotations
5 from abc import ABC, abstractmethod
6 import concurrent.futures as fut
7 from collections import defaultdict
8 from dataclasses import dataclass
17 from typing import Any, Callable, Dict, List, Optional, Set
19 import cloudpickle # type: ignore
20 from overrides import overrides
22 from ansi import bg, fg, underline, reset
25 from exec_utils import run_silently, cmd_in_background
26 from decorator_utils import singleton
27 import histogram as hist
29 logger = logging.getLogger(__name__)
31 parser = config.add_commandline_args(
32 f"Executors ({__file__})",
33 "Args related to processing executors."
36 '--executors_threadpool_size',
39 help='Number of threads in the default threadpool, leave unset for default',
43 '--executors_processpool_size',
46 help='Number of processes in the default processpool, leave unset for default',
50 '--executors_schedule_remote_backups',
52 action=argparse_utils.ActionNoYes,
53 help='Should we schedule duplicative backup work if a remote bundle is slow',
56 '--executors_max_bundle_failures',
60 help='Maximum number of failures before giving up on a bundle',
63 RSYNC = 'rsync -q --no-motd -W --ignore-existing --timeout=60 --size-only -z'
64 SSH = 'ssh -oForwardX11=no'
67 def make_cloud_pickle(fun, *args, **kwargs):
68 logger.debug(f"Making cloudpickled bundle at {fun.__name__}")
69 return cloudpickle.dumps((fun, args, kwargs))
72 class BaseExecutor(ABC):
73 def __init__(self, *, title=''):
76 self.histogram = hist.SimpleHistogram(
77 hist.SimpleHistogram.n_evenly_spaced_buckets(
86 **kwargs) -> fut.Future:
91 wait: bool = True) -> None:
94 def adjust_task_count(self, delta: int) -> None:
95 self.task_count += delta
96 logger.debug(f'Executor current task count is {self.task_count}')
99 class ThreadExecutor(BaseExecutor):
101 max_workers: Optional[int] = None):
104 if max_workers is not None:
105 workers = max_workers
106 elif 'executors_threadpool_size' in config.config:
107 workers = config.config['executors_threadpool_size']
108 logger.debug(f'Creating threadpool executor with {workers} workers')
109 self._thread_pool_executor = fut.ThreadPoolExecutor(
111 thread_name_prefix="thread_executor_helper"
114 def run_local_bundle(self, fun, *args, **kwargs):
115 logger.debug(f"Running local bundle at {fun.__name__}")
117 result = fun(*args, **kwargs)
119 self.adjust_task_count(-1)
120 duration = end - start
121 logger.debug(f"{fun.__name__} finished; used {duration:.1f}s")
122 self.histogram.add_item(duration)
129 **kwargs) -> fut.Future:
130 self.adjust_task_count(+1)
132 newargs.append(function)
135 return self._thread_pool_executor.submit(
136 self.run_local_bundle,
142 wait = True) -> None:
143 logger.debug(f'Shutting down threadpool executor {self.title}')
144 print(self.histogram)
145 self._thread_pool_executor.shutdown(wait)
148 class ProcessExecutor(BaseExecutor):
153 if max_workers is not None:
154 workers = max_workers
155 elif 'executors_processpool_size' in config.config:
156 workers = config.config['executors_processpool_size']
157 logger.debug(f'Creating processpool executor with {workers} workers.')
158 self._process_executor = fut.ProcessPoolExecutor(
162 def run_cloud_pickle(self, pickle):
163 fun, args, kwargs = cloudpickle.loads(pickle)
164 logger.debug(f"Running pickled bundle at {fun.__name__}")
165 result = fun(*args, **kwargs)
166 self.adjust_task_count(-1)
173 **kwargs) -> fut.Future:
175 self.adjust_task_count(+1)
176 pickle = make_cloud_pickle(function, *args, **kwargs)
177 result = self._process_executor.submit(
178 self.run_cloud_pickle,
181 result.add_done_callback(
182 lambda _: self.histogram.add_item(
189 def shutdown(self, wait=True) -> None:
190 logger.debug(f'Shutting down processpool executor {self.title}')
191 self._process_executor.shutdown(wait)
192 print(self.histogram)
194 def __getstate__(self):
195 state = self.__dict__.copy()
196 state['_process_executor'] = None
201 class RemoteWorkerRecord:
208 return hash((self.username, self.machine))
211 return f'{self.username}@{self.machine}'
219 worker: Optional[RemoteWorkerRecord]
220 username: Optional[str]
221 machine: Optional[str]
230 src_bundle: BundleDetails
231 is_cancelled: threading.Event
233 backup_bundles: Optional[List[BundleDetails]]
237 class RemoteExecutorStatus:
238 def __init__(self, total_worker_count: int) -> None:
239 self.worker_count = total_worker_count
240 self.known_workers: Set[RemoteWorkerRecord] = set()
241 self.start_per_bundle: Dict[str, float] = defaultdict(float)
242 self.end_per_bundle: Dict[str, float] = defaultdict(float)
243 self.finished_bundle_timings_per_worker: Dict[
247 self.in_flight_bundles_by_worker: Dict[
251 self.bundle_details_by_uuid: Dict[str, BundleDetails] = {}
252 self.finished_bundle_timings: List[float] = []
253 self.last_periodic_dump: Optional[float] = None
254 self.total_bundles_submitted = 0
256 # Protects reads and modification using self. Also used
257 # as a memory fence for modifications to bundle.
258 self.lock = threading.Lock()
260 def record_acquire_worker(
262 worker: RemoteWorkerRecord,
266 self.record_acquire_worker_already_locked(
271 def record_acquire_worker_already_locked(
273 worker: RemoteWorkerRecord,
276 assert self.lock.locked()
277 self.known_workers.add(worker)
278 self.start_per_bundle[uuid] = None
279 x = self.in_flight_bundles_by_worker.get(worker, set())
281 self.in_flight_bundles_by_worker[worker] = x
283 def record_bundle_details(
285 details: BundleDetails) -> None:
287 self.record_bundle_details_already_locked(details)
289 def record_bundle_details_already_locked(
291 details: BundleDetails) -> None:
292 assert self.lock.locked()
293 self.bundle_details_by_uuid[details.uuid] = details
295 def record_release_worker(
297 worker: RemoteWorkerRecord,
302 self.record_release_worker_already_locked(
303 worker, uuid, was_cancelled
306 def record_release_worker_already_locked(
308 worker: RemoteWorkerRecord,
312 assert self.lock.locked()
314 self.end_per_bundle[uuid] = ts
315 self.in_flight_bundles_by_worker[worker].remove(uuid)
316 if not was_cancelled:
317 bundle_latency = ts - self.start_per_bundle[uuid]
318 x = self.finished_bundle_timings_per_worker.get(worker, list())
319 x.append(bundle_latency)
320 self.finished_bundle_timings_per_worker[worker] = x
321 self.finished_bundle_timings.append(bundle_latency)
323 def record_processing_began(self, uuid: str):
325 self.start_per_bundle[uuid] = time.time()
327 def total_in_flight(self) -> int:
328 assert self.lock.locked()
330 for worker in self.known_workers:
331 total_in_flight += len(self.in_flight_bundles_by_worker[worker])
332 return total_in_flight
334 def total_idle(self) -> int:
335 assert self.lock.locked()
336 return self.worker_count - self.total_in_flight()
339 assert self.lock.locked()
341 total_finished = len(self.finished_bundle_timings)
342 total_in_flight = self.total_in_flight()
343 ret = f'\n\n{underline()}Remote Executor Pool Status{reset()}: '
345 if len(self.finished_bundle_timings) > 1:
346 qall = numpy.quantile(self.finished_bundle_timings, [0.5, 0.95])
348 f'⏱=∀p50:{qall[0]:.1f}s, ∀p95:{qall[1]:.1f}s, '
349 f'✅={total_finished}/{self.total_bundles_submitted}, '
350 f'💻n={total_in_flight}/{self.worker_count}\n'
354 f' ✅={total_finished}/{self.total_bundles_submitted}, '
355 f'💻n={total_in_flight}/{self.worker_count}\n'
358 for worker in self.known_workers:
359 ret += f' {fg("lightning yellow")}{worker.machine}{reset()}: '
360 timings = self.finished_bundle_timings_per_worker.get(worker, [])
364 qworker = numpy.quantile(timings, [0.5, 0.95])
365 ret += f' 💻p50: {qworker[0]:.1f}s, 💻p95: {qworker[1]:.1f}s\n'
369 ret += f' ...finished {count} total bundle(s) so far\n'
370 in_flight = len(self.in_flight_bundles_by_worker[worker])
372 ret += f' ...{in_flight} bundles currently in flight:\n'
373 for bundle_uuid in self.in_flight_bundles_by_worker[worker]:
374 details = self.bundle_details_by_uuid.get(
378 pid = str(details.pid) if details is not None else "TBD"
379 if self.start_per_bundle[bundle_uuid] is not None:
380 sec = ts - self.start_per_bundle[bundle_uuid]
381 ret += f' (pid={pid}): {bundle_uuid} for {sec:.1f}s so far '
383 ret += f' {bundle_uuid} setting up / copying data...'
386 if qworker is not None:
388 ret += f'{bg("red")}>💻p95{reset()} '
389 elif sec > qworker[0]:
390 ret += f'{fg("red")}>💻p50{reset()} '
392 if sec > qall[1] * 1.5:
393 ret += f'{bg("red")}!!!{reset()}'
394 if details is not None:
395 logger.debug(f'Flagging {details.uuid} for another backup')
396 details.super_slow = True
398 ret += f'{bg("red")}>∀p95{reset()} '
399 if details is not None:
400 logger.debug(f'Flagging {details.uuid} for a backup')
401 details.too_slow = True
403 ret += f'{fg("red")}>∀p50{reset()}'
407 def periodic_dump(self, total_bundles_submitted: int) -> None:
408 assert self.lock.locked()
409 self.total_bundles_submitted = total_bundles_submitted
412 self.last_periodic_dump is None
413 or ts - self.last_periodic_dump > 5.0
416 self.last_periodic_dump = ts
419 class RemoteWorkerSelectionPolicy(ABC):
420 def register_worker_pool(self, workers):
422 self.workers = workers
425 def is_worker_available(self) -> bool:
431 machine_to_avoid = None
432 ) -> Optional[RemoteWorkerRecord]:
436 class WeightedRandomRemoteWorkerSelectionPolicy(RemoteWorkerSelectionPolicy):
437 def is_worker_available(self) -> bool:
438 for worker in self.workers:
445 machine_to_avoid = None
446 ) -> Optional[RemoteWorkerRecord]:
448 for worker in self.workers:
449 for x in range(0, worker.count):
450 for y in range(0, worker.weight):
451 grabbag.append(worker)
453 for _ in range(0, 5):
454 random.shuffle(grabbag)
456 if worker.machine != machine_to_avoid or _ > 2:
459 logger.debug(f'Selected worker {worker}')
461 logger.warning("Couldn't find a worker; go fish.")
465 class RoundRobinRemoteWorkerSelectionPolicy(RemoteWorkerSelectionPolicy):
466 def __init__(self) -> None:
469 def is_worker_available(self) -> bool:
470 for worker in self.workers:
477 machine_to_avoid: str = None
478 ) -> Optional[RemoteWorkerRecord]:
481 worker = self.workers[x]
485 if x >= len(self.workers):
488 logger.debug(f'Selected worker {worker}')
491 if x >= len(self.workers):
494 logger.warning("Couldn't find a worker; go fish.")
498 class RemoteExecutor(BaseExecutor):
500 workers: List[RemoteWorkerRecord],
501 policy: RemoteWorkerSelectionPolicy) -> None:
503 self.workers = workers
505 self.worker_count = 0
506 for worker in self.workers:
507 self.worker_count += worker.count
508 if self.worker_count <= 0:
509 msg = f"We need somewhere to schedule work; count was {self.worker_count}"
512 self.policy.register_worker_pool(self.workers)
513 self.cv = threading.Condition()
514 logger.debug(f'Creating {self.worker_count} local threads, one per remote worker.')
515 self._helper_executor = fut.ThreadPoolExecutor(
516 thread_name_prefix="remote_executor_helper",
517 max_workers=self.worker_count,
519 self.status = RemoteExecutorStatus(self.worker_count)
520 self.total_bundles_submitted = 0
522 f'Creating remote processpool with {self.worker_count} remote worker threads.'
525 def is_worker_available(self) -> bool:
526 return self.policy.is_worker_available()
530 machine_to_avoid: str = None
531 ) -> Optional[RemoteWorkerRecord]:
532 return self.policy.acquire_worker(machine_to_avoid)
534 def find_available_worker_or_block(
536 machine_to_avoid: str = None
537 ) -> RemoteWorkerRecord:
539 while not self.is_worker_available():
541 worker = self.acquire_worker(machine_to_avoid)
542 if worker is not None:
544 msg = "We should never reach this point in the code"
548 def release_worker(self, worker: RemoteWorkerRecord) -> None:
549 logger.debug(f'Released worker {worker}')
554 def heartbeat(self) -> None:
555 with self.status.lock:
556 # Regular progress report
557 self.status.periodic_dump(self.total_bundles_submitted)
559 # Look for bundles to reschedule
560 num_done = len(self.status.finished_bundle_timings)
561 if num_done > 7 or (num_done > 5 and self.is_worker_available()):
562 for worker, bundle_uuids in self.status.in_flight_bundles_by_worker.items():
563 for uuid in bundle_uuids:
564 bundle = self.status.bundle_details_by_uuid.get(uuid, None)
566 bundle is not None and
568 bundle.src_bundle is None and
569 config.config['executors_schedule_remote_backups']
571 self.consider_backup_for_bundle(bundle)
573 def consider_backup_for_bundle(self, bundle: BundleDetails) -> None:
574 assert self.status.lock.locked()
577 and len(bundle.backup_bundles) == 0 # one backup per
579 msg = f"*** Rescheduling {bundle.pid}/{bundle.uuid} ***"
581 self.schedule_backup_for_bundle(bundle)
585 and len(bundle.backup_bundles) < 2 # two backups in dire situations
586 and self.status.total_idle() > 4
588 msg = f"*** Rescheduling {bundle.pid}/{bundle.uuid} ***"
590 self.schedule_backup_for_bundle(bundle)
593 def check_if_cancelled(self, bundle: BundleDetails) -> bool:
594 with self.status.lock:
595 if bundle.is_cancelled.wait(timeout=0.0):
596 logger.debug(f'Bundle {bundle.uuid} is cancelled, bail out.')
597 bundle.was_cancelled = True
601 def launch(self, bundle: BundleDetails, override_avoid_machine=None) -> Any:
602 """Find a worker for bundle or block until one is available."""
603 self.adjust_task_count(+1)
605 hostname = bundle.hostname
606 avoid_machine = override_avoid_machine
607 is_original = bundle.src_bundle is None
609 # Try not to schedule a backup on the same host as the original.
610 if avoid_machine is None and bundle.src_bundle is not None:
611 avoid_machine = bundle.src_bundle.machine
613 while worker is None:
614 worker = self.find_available_worker_or_block(avoid_machine)
616 # Ok, found a worker.
617 bundle.worker = worker
618 machine = bundle.machine = worker.machine
619 username = bundle.username = worker.username
621 self.status.record_acquire_worker(worker, uuid)
622 logger.debug(f'{uuid}/{fname}: Running bundle on {worker}...')
624 # Before we do any work, make sure the bundle is still viable.
625 if self.check_if_cancelled(bundle):
627 return self.post_launch_work(bundle)
628 except Exception as e:
631 f'{uuid}/{fname}: bundle says it\'s cancelled upfront but no results?!'
633 assert bundle.worker is not None
634 self.status.record_release_worker(
639 self.release_worker(bundle.worker)
640 self.adjust_task_count(-1)
642 # Weird. We are the original owner of this
643 # bundle. For it to have been cancelled, a backup
644 # must have already started and completed before
645 # we even for started. Moreover, the backup says
646 # it is done but we can't find the results it
647 # should have copied over. Reschedule the whole
649 return self.emergency_retry_nasty_bundle(bundle)
651 # Expected(?). We're a backup and our bundle is
652 # cancelled before we even got started. Something
653 # went bad in post_launch_work (I acutually don't
654 # see what?) but probably not worth worrying
658 # Send input code / data to worker machine if it's not local.
659 if hostname not in machine:
661 cmd = f'{RSYNC} {bundle.code_file} {username}@{machine}:{bundle.code_file}'
662 start_ts = time.time()
663 logger.info(f"{uuid}/{fname}: Copying work to {worker} via {cmd}.")
665 xfer_latency = time.time() - start_ts
666 logger.info(f"{uuid}/{fname}: Copying done in {xfer_latency}s.")
667 except Exception as e:
670 f'{uuid}/{fname}: failed to send instructions to worker machine?!?'
672 assert bundle.worker is not None
673 self.status.record_release_worker(
678 self.release_worker(bundle.worker)
679 self.adjust_task_count(-1)
681 # Weird. We tried to copy the code to the worker and it failed...
682 # And we're the original bundle. We have to retry.
683 return self.emergency_retry_nasty_bundle(bundle)
685 # This is actually expected; we're a backup.
686 # There's a race condition where someone else
687 # already finished the work and removed the source
688 # code file before we could copy it. No biggie.
691 # Kick off the work. Note that if this fails we let
692 # wait_for_process deal with it.
693 self.status.record_processing_began(uuid)
694 cmd = (f'{SSH} {bundle.username}@{bundle.machine} '
695 f'"source py39-venv/bin/activate &&'
696 f' /home/scott/lib/python_modules/remote_worker.py'
697 f' --code_file {bundle.code_file} --result_file {bundle.result_file}"')
698 logger.debug(f'{uuid}/{fname}: Executing {cmd} in the background to kick off work...')
699 p = cmd_in_background(cmd, silent=True)
700 bundle.pid = pid = p.pid
701 logger.debug(f'{uuid}/{fname}: Local ssh process pid={pid}; remote worker is {machine}.')
702 return self.wait_for_process(p, bundle, 0)
704 def wait_for_process(self, p: subprocess.Popen, bundle: BundleDetails, depth: int) -> Any:
706 machine = bundle.machine
711 f"I've gotten repeated errors waiting on this bundle; giving up on pid={pid}."
714 self.status.record_release_worker(
719 self.release_worker(bundle.worker)
720 self.adjust_task_count(-1)
721 return self.emergency_retry_nasty_bundle(bundle)
723 # Spin until either the ssh job we scheduled finishes the
724 # bundle or some backup worker signals that they finished it
729 except subprocess.TimeoutExpired:
731 if self.check_if_cancelled(bundle):
733 f'{uuid}/{fname}: another worker finished bundle, checking it out...'
738 f"{uuid}/{fname}: pid {pid} ({machine}) our ssh finished, checking it out..."
743 # If we get here we believe the bundle is done; either the ssh
744 # subprocess finished (hopefully successfully) or we noticed
745 # that some other worker seems to have completed the bundle
746 # and we're bailing out.
748 ret = self.post_launch_work(bundle)
749 if ret is not None and p is not None:
753 # Something went wrong; e.g. we could not copy the results
754 # back, cleanup after ourselves on the remote machine, or
755 # unpickle the results we got from the remove machine. If we
756 # still have an active ssh subprocess, keep waiting on it.
757 # Otherwise, time for an emergency reschedule.
758 except Exception as e:
760 logger.error(f'{uuid}/{fname}: Something unexpected just happened...')
763 f"{uuid}/{fname}: Failed to wrap up \"done\" bundle, re-waiting on active ssh."
765 return self.wait_for_process(p, bundle, depth + 1)
767 self.status.record_release_worker(
772 self.release_worker(bundle.worker)
773 self.adjust_task_count(-1)
774 return self.emergency_retry_nasty_bundle(bundle)
776 def post_launch_work(self, bundle: BundleDetails) -> Any:
777 with self.status.lock:
778 is_original = bundle.src_bundle is None
779 was_cancelled = bundle.was_cancelled
780 username = bundle.username
781 machine = bundle.machine
782 result_file = bundle.result_file
783 code_file = bundle.code_file
787 # Whether original or backup, if we finished first we must
788 # fetch the results if the computation happened on a
790 bundle.end_ts = time.time()
791 if not was_cancelled:
792 assert bundle.machine is not None
793 if bundle.hostname not in bundle.machine:
794 cmd = f'{RSYNC} {username}@{machine}:{result_file} {result_file} 2>/dev/null'
796 f"{uuid}/{fname}: Fetching results from {username}@{machine} via {cmd}"
799 # If either of these throw they are handled in
802 run_silently(f'{SSH} {username}@{machine}'
803 f' "/bin/rm -f {code_file} {result_file}"')
804 dur = bundle.end_ts - bundle.start_ts
805 self.histogram.add_item(dur)
807 # Only the original worker should unpickle the file contents
808 # though since it's the only one whose result matters. The
809 # original is also the only job that may delete result_file
810 # from disk. Note that the original may have been cancelled
811 # if one of the backups finished first; it still must read the
814 logger.debug(f"{uuid}/{fname}: Unpickling {result_file}.")
816 with open(f'{result_file}', 'rb') as rb:
817 serialized = rb.read()
818 result = cloudpickle.loads(serialized)
819 except Exception as e:
820 msg = f'Failed to load {result_file}, this is bad news.'
822 self.status.record_release_worker(
827 self.release_worker(bundle.worker)
829 # Re-raise the exception; the code in wait_for_process may
830 # decide to emergency_retry_nasty_bundle here.
834 f'Removing local (master) {code_file} and {result_file}.'
836 os.remove(f'{result_file}')
837 os.remove(f'{code_file}')
839 # Notify any backups that the original is done so they
840 # should stop ASAP. Do this whether or not we
841 # finished first since there could be more than one
843 if bundle.backup_bundles is not None:
844 for backup in bundle.backup_bundles:
846 f'{uuid}/{fname}: Notifying backup {backup.uuid} that it\'s cancelled'
848 backup.is_cancelled.set()
850 # This is a backup job and, by now, we have already fetched
851 # the bundle results.
853 # Backup results don't matter, they just need to leave the
854 # result file in the right place for their originals to
855 # read/unpickle later.
858 # Tell the original to stop if we finished first.
859 if not was_cancelled:
861 f'{uuid}/{fname}: Notifying original {bundle.src_bundle.uuid} we beat them to it.'
863 bundle.src_bundle.is_cancelled.set()
865 assert bundle.worker is not None
866 self.status.record_release_worker(
871 self.release_worker(bundle.worker)
872 self.adjust_task_count(-1)
875 def create_original_bundle(self, pickle, fname: str):
876 from string_utils import generate_uuid
877 uuid = generate_uuid(as_hex=True)
878 code_file = f'/tmp/{uuid}.code.bin'
879 result_file = f'/tmp/{uuid}.result.bin'
881 logger.debug(f'Writing pickled code to {code_file}')
882 with open(f'{code_file}', 'wb') as wb:
885 bundle = BundleDetails(
886 pickled_code = pickle,
892 hostname = platform.node(),
893 code_file = code_file,
894 result_file = result_file,
896 start_ts = time.time(),
901 is_cancelled = threading.Event(),
902 was_cancelled = False,
906 self.status.record_bundle_details(bundle)
907 logger.debug(f'{uuid}/{fname}: Created original bundle')
910 def create_backup_bundle(self, src_bundle: BundleDetails):
911 assert src_bundle.backup_bundles is not None
912 n = len(src_bundle.backup_bundles)
913 uuid = src_bundle.uuid + f'_backup#{n}'
915 backup_bundle = BundleDetails(
916 pickled_code = src_bundle.pickled_code,
918 fname = src_bundle.fname,
922 hostname = src_bundle.hostname,
923 code_file = src_bundle.code_file,
924 result_file = src_bundle.result_file,
926 start_ts = time.time(),
930 src_bundle = src_bundle,
931 is_cancelled = threading.Event(),
932 was_cancelled = False,
933 backup_bundles = None, # backup backups not allowed
936 src_bundle.backup_bundles.append(backup_bundle)
937 self.status.record_bundle_details_already_locked(backup_bundle)
938 logger.debug(f'{uuid}/{src_bundle.fname}: Created backup bundle')
941 def schedule_backup_for_bundle(self,
942 src_bundle: BundleDetails):
943 assert self.status.lock.locked()
944 backup_bundle = self.create_backup_bundle(src_bundle)
946 f'{backup_bundle.uuid}/{backup_bundle.fname}: Scheduling backup for execution...'
948 self._helper_executor.submit(self.launch, backup_bundle)
950 # Results from backups don't matter; if they finish first
951 # they will move the result_file to this machine and let
952 # the original pick them up and unpickle them.
954 def emergency_retry_nasty_bundle(self, bundle: BundleDetails) -> fut.Future:
956 is_original = bundle.src_bundle is None
958 avoid_last_machine = bundle.machine
959 bundle.machine = None
960 bundle.username = None
961 bundle.failure_count += 1
967 if bundle.failure_count > retry_limit:
969 f'{uuid}: Tried this bundle too many times already ({retry_limit}x); giving up.'
973 f'{uuid}: This is the original of the bundle; results will be incomplete.'
976 logger.error(f'{uuid}: At least it\'s only a backup; better luck with the others.')
980 f'>>> Emergency rescheduling {uuid} because of unexected errors (wtf?!) <<<'
982 return self.launch(bundle, avoid_last_machine)
988 **kwargs) -> fut.Future:
989 pickle = make_cloud_pickle(function, *args, **kwargs)
990 bundle = self.create_original_bundle(pickle, function.__name__)
991 self.total_bundles_submitted += 1
992 return self._helper_executor.submit(self.launch, bundle)
995 def shutdown(self, wait=True) -> None:
996 self._helper_executor.shutdown(wait)
997 logging.debug(f'Shutting down RemoteExecutor {self.title}')
998 print(self.histogram)
1002 class DefaultExecutors(object):
1004 self.thread_executor: Optional[ThreadExecutor] = None
1005 self.process_executor: Optional[ProcessExecutor] = None
1006 self.remote_executor: Optional[RemoteExecutor] = None
1008 def ping(self, host) -> bool:
1009 logger.debug(f'RUN> ping -c 1 {host}')
1010 command = ['ping', '-c', '1', host]
1011 return subprocess.call(
1013 stdout=subprocess.DEVNULL,
1014 stderr=subprocess.DEVNULL,
1017 def thread_pool(self) -> ThreadExecutor:
1018 if self.thread_executor is None:
1019 self.thread_executor = ThreadExecutor()
1020 return self.thread_executor
1022 def process_pool(self) -> ProcessExecutor:
1023 if self.process_executor is None:
1024 self.process_executor = ProcessExecutor()
1025 return self.process_executor
1027 def remote_pool(self) -> RemoteExecutor:
1028 if self.remote_executor is None:
1029 logger.info('Looking for some helper machines...')
1030 pool: List[RemoteWorkerRecord] = []
1031 if self.ping('cheetah.house'):
1032 logger.info('Found cheetah.house')
1036 machine = 'cheetah.house',
1041 if self.ping('video.house'):
1042 logger.info('Found video.house')
1046 machine = 'video.house',
1051 if self.ping('wannabe.house'):
1052 logger.info('Found wannabe.house')
1056 machine = 'wannabe.house',
1061 if self.ping('meerkat.cabin'):
1062 logger.info('Found meerkat.cabin')
1066 machine = 'meerkat.cabin',
1071 if self.ping('backup.house'):
1072 logger.info('Found backup.house')
1076 machine = 'backup.house',
1081 if self.ping('kiosk.house'):
1082 logger.info('Found kiosk.house')
1086 machine = 'kiosk.house',
1091 if self.ping('puma.cabin'):
1092 logger.info('Found puma.cabin')
1096 machine = 'puma.cabin',
1102 # The controller machine has a lot to do; go easy on it.
1104 if record.machine == platform.node() and record.count > 1:
1105 logger.info(f'Reducing workload for {record.machine}.')
1108 policy = WeightedRandomRemoteWorkerSelectionPolicy()
1109 policy.register_worker_pool(pool)
1110 self.remote_executor = RemoteExecutor(pool, policy)
1111 return self.remote_executor