From eedcbd4f64af13ec2098508c3d839a60f7e9ffce Mon Sep 17 00:00:00 2001 From: Scott Date: Sun, 23 Jan 2022 12:43:46 -0800 Subject: [PATCH] Clean up the remote executor stuff and create a dedicated heartbeat thread. Fix the parallelize integration test. Other changes that depended on these fixes. --- executors.py | 362 ++++++++++++++++++++----------------- file_utils.py | 14 +- ml/model_trainer.py | 3 + tests/parallelize_itest.py | 22 +-- 4 files changed, 210 insertions(+), 191 deletions(-) diff --git a/executors.py b/executors.py index 2829c69..c91c2a6 100644 --- a/executors.py +++ b/executors.py @@ -26,6 +26,8 @@ import config from decorator_utils import singleton from exec_utils import run_silently, cmd_in_background, cmd_with_timeout import histogram as hist +from thread_utils import background_thread + logger = logging.getLogger(__name__) @@ -64,6 +66,7 @@ parser.add_argument( SSH = '/usr/bin/ssh -oForwardX11=no' SCP = '/usr/bin/scp' + def make_cloud_pickle(fun, *args, **kwargs): logger.debug(f"Making cloudpickled bundle at {fun.__name__}") return cloudpickle.dumps((fun, args, kwargs)) @@ -556,6 +559,119 @@ class RemoteExecutor(BaseExecutor): self.total_bundles_submitted = 0 self.backup_lock = threading.Lock() self.last_backup = None + (self.heartbeat_thread, self.heartbeat_stop_event) = self.run_periodic_heartbeat() + + @background_thread + def run_periodic_heartbeat(self, stop_event) -> None: + while not stop_event.is_set(): + time.sleep(5.0) + logger.debug('Running periodic heartbeat code...') + self.heartbeat() + logger.debug('Periodic heartbeat thread shutting down.') + + def heartbeat(self) -> None: + with self.status.lock: + # Dump regular progress report + self.status.periodic_dump(self.total_bundles_submitted) + + # Look for bundles to reschedule via executor.submit + if config.config['executors_schedule_remote_backups']: + self.maybe_schedule_backup_bundles() + + def maybe_schedule_backup_bundles(self): + assert self.status.lock.locked() + num_done = len(self.status.finished_bundle_timings) + num_idle_workers = self.worker_count - self.task_count + now = time.time() + if ( + num_done > 2 + and num_idle_workers > 1 + and (self.last_backup is None or (now - self.last_backup > 6.0)) + and self.backup_lock.acquire(blocking=False) + ): + try: + assert self.backup_lock.locked() + + bundle_to_backup = None + best_score = None + for worker, bundle_uuids in self.status.in_flight_bundles_by_worker.items(): + + # Prefer to schedule backups of bundles running on + # slower machines. + base_score = 0 + for record in self.workers: + if worker.machine == record.machine: + base_score = float(record.weight) + base_score = 1.0 / base_score + base_score *= 200.0 + base_score = int(base_score) + break + + for uuid in bundle_uuids: + bundle = self.status.bundle_details_by_uuid.get(uuid, None) + if ( + bundle is not None + and bundle.src_bundle is None + and bundle.backup_bundles is not None + ): + score = base_score + + # Schedule backups of bundles running + # longer; especially those that are + # unexpectedly slow. + start_ts = self.status.start_per_bundle[uuid] + if start_ts is not None: + runtime = now - start_ts + score += runtime + logger.debug(f'score[{bundle}] => {score} # latency boost') + + if bundle.slower_than_local_p95: + score += runtime / 2 + logger.debug(f'score[{bundle}] => {score} # >worker p95') + + if bundle.slower_than_global_p95: + score += runtime / 4 + logger.debug(f'score[{bundle}] => {score} # >global p95') + + # Prefer backups of bundles that don't + # have backups already. + backup_count = len(bundle.backup_bundles) + if backup_count == 0: + score *= 2 + elif backup_count == 1: + score /= 2 + elif backup_count == 2: + score /= 8 + else: + score = 0 + logger.debug( + f'score[{bundle}] => {score} # {backup_count} dup backup factor' + ) + + if ( + score != 0 + and (best_score is None or score > best_score) + ): + bundle_to_backup = bundle + assert bundle is not None + assert bundle.backup_bundles is not None + assert bundle.src_bundle is None + best_score = score + + # Note: this is all still happening on the heartbeat + # runner thread. That's ok because + # schedule_backup_for_bundle uses the executor to + # submit the bundle again which will cause it to be + # picked up by a worker thread and allow this thread + # to return to run future heartbeats. + if bundle_to_backup is not None: + self.last_backup = now + logger.info( + f'=====> SCHEDULING BACKUP {bundle_to_backup} (score={best_score:.1f}) <=====' + ) + self.schedule_backup_for_bundle(bundle_to_backup) + finally: + self.backup_lock.release() def is_worker_available(self) -> bool: return self.policy.is_worker_available() @@ -580,97 +696,24 @@ class RemoteExecutor(BaseExecutor): logger.critical(msg) raise Exception(msg) - def release_worker(self, worker: RemoteWorkerRecord) -> None: + def release_worker( + self, + bundle: BundleDetails, + *, + was_cancelled=True + ) -> None: + worker = bundle.worker + assert worker is not None logger.debug(f'Released worker {worker}') + self.status.record_release_worker( + worker, + bundle.uuid, + was_cancelled, + ) with self.cv: worker.count += 1 self.cv.notify() - - def heartbeat(self) -> None: - with self.status.lock: - # Regular progress report - self.status.periodic_dump(self.total_bundles_submitted) - - # Look for bundles to reschedule. - num_done = len(self.status.finished_bundle_timings) - num_idle_workers = self.worker_count - self.task_count - now = time.time() - if ( - config.config['executors_schedule_remote_backups'] - and num_done > 2 - and num_idle_workers > 1 - and (self.last_backup is None or (now - self.last_backup > 1.0)) - and self.backup_lock.acquire(blocking=False) - ): - try: - assert self.backup_lock.locked() - - bundle_to_backup = None - best_score = None - for worker, bundle_uuids in self.status.in_flight_bundles_by_worker.items(): - # Prefer to schedule backups of bundles on slower machines. - base_score = 0 - for record in self.workers: - if worker.machine == record.machine: - base_score = float(record.weight) - base_score = 1.0 / base_score - base_score *= 200.0 - base_score = int(base_score) - break - - for uuid in bundle_uuids: - bundle = self.status.bundle_details_by_uuid.get(uuid, None) - if ( - bundle is not None - and bundle.src_bundle is None - and bundle.backup_bundles is not None - ): - score = base_score - - # Schedule backups of bundles running longer; especially those - # that are unexpectedly slow. - start_ts = self.status.start_per_bundle[uuid] - if start_ts is not None: - runtime = now - start_ts - score += runtime - logger.debug(f'score[{bundle}] => {score} # latency boost') - - if bundle.slower_than_local_p95: - score += runtime / 2 - logger.debug(f'score[{bundle}] => {score} # >worker p95') - - if bundle.slower_than_global_p95: - score += runtime / 2 - logger.debug(f'score[{bundle}] => {score} # >global p95') - - # Prefer backups of bundles that don't have backups already. - backup_count = len(bundle.backup_bundles) - if backup_count == 0: - score *= 2 - elif backup_count == 1: - score /= 2 - elif backup_count == 2: - score /= 8 - else: - score = 0 - logger.debug(f'score[{bundle}] => {score} # {backup_count} dup backup factor') - - if ( - score != 0 - and (best_score is None or score > best_score) - ): - bundle_to_backup = bundle - assert bundle is not None - assert bundle.backup_bundles is not None - assert bundle.src_bundle is None - best_score = score - - if bundle_to_backup is not None: - self.last_backup = now - logger.info(f'=====> SCHEDULING BACKUP {bundle_to_backup} (score={best_score:.1f}) <=====') - self.schedule_backup_for_bundle(bundle_to_backup) - finally: - self.backup_lock.release() + self.adjust_task_count(-1) def check_if_cancelled(self, bundle: BundleDetails) -> bool: with self.status.lock: @@ -694,32 +737,27 @@ class RemoteExecutor(BaseExecutor): worker = None while worker is None: worker = self.find_available_worker_or_block(avoid_machine) + assert worker # Ok, found a worker. bundle.worker = worker machine = bundle.machine = worker.machine username = bundle.username = worker.username - self.status.record_acquire_worker(worker, uuid) logger.debug(f'{bundle}: Running bundle on {worker}...') # Before we do any work, make sure the bundle is still viable. + # It may have been some time between when it was submitted and + # now due to lack of worker availability and someone else may + # have already finished it. if self.check_if_cancelled(bundle): try: - return self.post_launch_work(bundle) + return self.process_work_result(bundle) except Exception as e: - logger.exception(e) - logger.error( + logger.warning( f'{bundle}: bundle says it\'s cancelled upfront but no results?!' ) - assert bundle.worker is not None - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - True, - ) - self.release_worker(bundle.worker) - self.adjust_task_count(-1) + self.release_worker(bundle) if is_original: # Weird. We are the original owner of this # bundle. For it to have been cancelled, a backup @@ -728,13 +766,20 @@ class RemoteExecutor(BaseExecutor): # it is done but we can't find the results it # should have copied over. Reschedule the whole # thing. + logger.exception(e) + logger.error( + f'{bundle}: We are the original owner thread and yet there are ' + + 'no results for this bundle. This is unexpected and bad.' + ) return self.emergency_retry_nasty_bundle(bundle) else: # Expected(?). We're a backup and our bundle is # cancelled before we even got started. Something - # went bad in post_launch_work (I acutually don't + # went bad in process_work_result (I acutually don't # see what?) but probably not worth worrying - # about. + # about. Let the original thread worry about + # either finding the results or complaining about + # it. return None # Send input code / data to worker machine if it's not local. @@ -745,24 +790,17 @@ class RemoteExecutor(BaseExecutor): logger.info(f"{bundle}: Copying work to {worker} via {cmd}.") run_silently(cmd) xfer_latency = time.time() - start_ts - logger.info(f"{bundle}: Copying done to {worker} in {xfer_latency:.1f}s.") + logger.info(f"{bundle}: Copying to {worker} took {xfer_latency:.1f}s.") except Exception as e: - assert bundle.worker is not None - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - True, - ) - self.release_worker(bundle.worker) - self.adjust_task_count(-1) + self.release_worker(bundle) if is_original: # Weird. We tried to copy the code to the worker and it failed... # And we're the original bundle. We have to retry. logger.exception(e) logger.error( - f'{bundle}: Failed to send instructions to the worker machine?! ' + - 'This is not expected; we\'re the original bundle so this shouldn\'t ' + - 'be a race condition. Attempting an emergency retry...' + f"{bundle}: Failed to send instructions to the worker machine?! " + + "This is not expected; we\'re the original bundle so this shouldn\'t " + + "be a race condition. Attempting an emergency retry..." ) return self.emergency_retry_nasty_bundle(bundle) else: @@ -785,8 +823,8 @@ class RemoteExecutor(BaseExecutor): f' --code_file {bundle.code_file} --result_file {bundle.result_file}"') logger.debug(f'{bundle}: Executing {cmd} in the background to kick off work...') p = cmd_in_background(cmd, silent=True) - bundle.pid = pid = p.pid - logger.debug(f'{bundle}: Local ssh process pid={pid}; remote worker is {machine}.') + bundle.pid = p.pid + logger.debug(f'{bundle}: Local ssh process pid={p.pid}; remote worker is {machine}.') return self.wait_for_process(p, bundle, 0) def wait_for_process(self, p: subprocess.Popen, bundle: BundleDetails, depth: int) -> Any: @@ -797,13 +835,7 @@ class RemoteExecutor(BaseExecutor): f"I've gotten repeated errors waiting on this bundle; giving up on pid={pid}." ) p.terminate() - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - True, - ) - self.release_worker(bundle.worker) - self.adjust_task_count(-1) + self.release_worker(bundle) return self.emergency_retry_nasty_bundle(bundle) # Spin until either the ssh job we scheduled finishes the @@ -813,7 +845,6 @@ class RemoteExecutor(BaseExecutor): try: p.wait(timeout=0.25) except subprocess.TimeoutExpired: - self.heartbeat() if self.check_if_cancelled(bundle): logger.info( f'{bundle}: another worker finished bundle, checking it out...' @@ -821,7 +852,7 @@ class RemoteExecutor(BaseExecutor): break else: logger.info( - f"{bundle}: pid {pid} ({machine}) our ssh finished, checking it out..." + f"{bundle}: pid {pid} ({machine}) is finished!" ) p = None break @@ -831,7 +862,7 @@ class RemoteExecutor(BaseExecutor): # that some other worker seems to have completed the bundle # and we're bailing out. try: - ret = self.post_launch_work(bundle) + ret = self.process_work_result(bundle) if ret is not None and p is not None: p.terminate() return ret @@ -849,16 +880,10 @@ class RemoteExecutor(BaseExecutor): logger.warning(msg) return self.wait_for_process(p, bundle, depth + 1) else: - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - True, - ) - self.release_worker(bundle.worker) - self.adjust_task_count(-1) + self.release_worker(bundle) return self.emergency_retry_nasty_bundle(bundle) - def post_launch_work(self, bundle: BundleDetails) -> Any: + def process_work_result(self, bundle: BundleDetails) -> Any: with self.status.lock: is_original = bundle.src_bundle is None was_cancelled = bundle.was_cancelled @@ -876,14 +901,25 @@ class RemoteExecutor(BaseExecutor): if bundle.hostname not in bundle.machine: cmd = f'{SCP} {username}@{machine}:{result_file} {result_file} 2>/dev/null' logger.info( - f"{bundle}: Fetching results from {username}@{machine} via {cmd}" + f"{bundle}: Fetching results back from {username}@{machine} via {cmd}" ) # If either of these throw they are handled in # wait_for_process. - run_silently(cmd) + attempts = 0 + while True: + try: + run_silently(cmd) + except Exception as e: + attempts += 1 + if attempts >= 3: + raise(e) + else: + break + run_silently(f'{SSH} {username}@{machine}' f' "/bin/rm -f {code_file} {result_file}"') + logger.debug(f'Fetching results back took {time.time() - bundle.end_ts:.1f}s.') dur = bundle.end_ts - bundle.start_ts self.histogram.add_item(dur) @@ -900,19 +936,14 @@ class RemoteExecutor(BaseExecutor): serialized = rb.read() result = cloudpickle.loads(serialized) except Exception as e: - msg = f'Failed to load {result_file}, this is bad news.' + logger.exception(e) + msg = f'Failed to load {result_file}... this is bad news.' logger.critical(msg) - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - True, - ) - self.release_worker(bundle.worker) + self.release_worker(bundle) # Re-raise the exception; the code in wait_for_process may # decide to emergency_retry_nasty_bundle here. raise Exception(e) - logger.debug( f'Removing local (master) {code_file} and {result_file}.' ) @@ -944,15 +975,7 @@ class RemoteExecutor(BaseExecutor): f'{bundle}: Notifying original {bundle.src_bundle.uuid} we beat them to it.' ) bundle.src_bundle.is_cancelled.set() - - assert bundle.worker is not None - self.status.record_release_worker( - bundle.worker, - bundle.uuid, - was_cancelled, - ) - self.release_worker(bundle.worker) - self.adjust_task_count(-1) + self.release_worker(bundle, was_cancelled=was_cancelled) return result def create_original_bundle(self, pickle, fname: str): @@ -1076,8 +1099,10 @@ class RemoteExecutor(BaseExecutor): @overrides def shutdown(self, wait=True) -> None: - self._helper_executor.shutdown(wait) logging.debug(f'Shutting down RemoteExecutor {self.title}') + self.heartbeat_stop_event.set() + self.heartbeat_thread.join() + self._helper_executor.shutdown(wait) print(self.histogram) @@ -1129,26 +1154,16 @@ class DefaultExecutors(object): RemoteWorkerRecord( username = 'scott', machine = 'meerkat.cabin', - weight = 5, + weight = 12, count = 2, ), ) - # if self.ping('kiosk.house'): - # logger.info('Found kiosk.house') - # pool.append( - # RemoteWorkerRecord( - # username = 'pi', - # machine = 'kiosk.house', - # weight = 1, - # count = 2, - # ), - # ) - if self.ping('hero.house'): - logger.info('Found hero.house') + if self.ping('wannabe.house'): + logger.info('Found wannabe.house') pool.append( RemoteWorkerRecord( username = 'scott', - machine = 'hero.house', + machine = 'wannabe.house', weight = 30, count = 10, ), @@ -1169,7 +1184,7 @@ class DefaultExecutors(object): RemoteWorkerRecord( username = 'scott', machine = 'backup.house', - weight = 3, + weight = 7, count = 2, ), ) @@ -1184,3 +1199,14 @@ class DefaultExecutors(object): policy.register_worker_pool(pool) self.remote_executor = RemoteExecutor(pool, policy) return self.remote_executor + + def shutdown(self) -> None: + if self.thread_executor is not None: + self.thread_executor.shutdown() + self.thread_executor = None + if self.process_executor is not None: + self.process_executor.shutdown() + self.process_executor = None + if self.remote_executor is not None: + self.remote_executor.shutdown() + self.remote_executor = None diff --git a/file_utils.py b/file_utils.py index e40a9f5..9fa8fd2 100644 --- a/file_utils.py +++ b/file_utils.py @@ -249,9 +249,7 @@ def is_same_file(file1: str, file2: str) -> bool: def get_file_raw_timestamps(filename: str) -> Optional[os.stat_result]: - """Stats the file and returns an os.stat_result or None on error. - - """ + """Stats the file and returns an os.stat_result or None on error.""" try: return os.stat(filename) except Exception as e: @@ -279,9 +277,7 @@ def get_file_raw_ctime(filename: str) -> Optional[float]: def get_file_md5(filename: str) -> str: - """Hashes filename's contents and returns an MD5. - - """ + """Hashes filename's contents and returns an MD5.""" file_hash = hashlib.md5() with open(filename, "rb") as f: chunk = f.read(8192) @@ -371,10 +367,9 @@ def get_file_mtime_timedelta(filename: str) -> Optional[datetime.timedelta]: return get_file_timestamp_timedelta(filename, lambda x: x.st_mtime) -def describe_file_timestamp( - filename: str, extractor, *, brief=False -) -> Optional[str]: +def describe_file_timestamp(filename: str, extractor, *, brief=False) -> Optional[str]: from datetime_utils import describe_duration, describe_duration_briefly + age = get_file_timestamp_age_seconds(filename, extractor) if age is None: return None @@ -451,4 +446,5 @@ class FileWriter(object): if __name__ == '__main__': import doctest + doctest.testmod() diff --git a/ml/model_trainer.py b/ml/model_trainer.py index 79ce706..041f0f8 100644 --- a/ml/model_trainer.py +++ b/ml/model_trainer.py @@ -22,6 +22,7 @@ from ansi import bold, reset import argparse_utils import config from decorator_utils import timed +import executors import parallelize as par logger = logging.getLogger(__file__) @@ -171,9 +172,11 @@ class TrainingBlueprint(ABC): ) if not self.spec.quiet: + executors.DefaultExecutors().shutdown() msg = f"Done training; best test set score was: {best_test_score:.1f}%" print(msg) logger.info(msg) + scaler_filename, model_filename, model_info_filename = ( self.maybe_persist_scaler_and_model( best_training_score, diff --git a/tests/parallelize_itest.py b/tests/parallelize_itest.py index 9d98710..11c5676 100755 --- a/tests/parallelize_itest.py +++ b/tests/parallelize_itest.py @@ -1,13 +1,11 @@ #!/usr/bin/env python3 -import random import sys import bootstrap import parallelize as p import decorator_utils import executors -import math_utils import smart_future @@ -28,12 +26,11 @@ def compute_factorial_process(n): @p.parallelize(method=p.Method.REMOTE) -def list_primes(n): - """Calculates sum of all primes below given integer n""" - ret = [] +def compute_factorial_remote(n): + total = 1 for x in range(2, n): - ret.append(math_utils.is_prime(x)) - return ret + total *= x + return total @decorator_utils.timed @@ -61,14 +58,11 @@ def test_process_parallelization() -> None: @decorator_utils.timed def test_remote_parallelization() -> None: - results = {} + results = [] for _ in range(50): - n = random.randint(0, 100000) - results[n] = list_primes(n) - tot = 0 - for _ in results[n]: - tot += _ - print(tot) + results.append(compute_factorial_remote(_)) + for result in smart_future.wait_any(results): + print(result) rexecutor = executors.DefaultExecutors().remote_pool() rexecutor.shutdown() -- 2.45.2