X-Git-Url: https://wannabe.guru.org/gitweb/?a=blobdiff_plain;f=executors.py;h=3cb0a916c080128e63a23600db76c07e93956ec9;hb=592f27f4743c0a66b6634692fea6c3eb8f252814;hp=ddd62f1b8a3998fa85a4fb456538eabd54e3d206;hpb=a5c4feeac240a446a38147130b51ad96a046f6e1;p=python_utils.git diff --git a/executors.py b/executors.py index ddd62f1..3cb0a91 100644 --- a/executors.py +++ b/executors.py @@ -210,6 +210,7 @@ class RemoteWorkerRecord: class BundleDetails: pickled_code: bytes uuid: str + fname: str worker: Optional[RemoteWorkerRecord] username: Optional[str] machine: Optional[str] @@ -475,6 +476,7 @@ class RemoteExecutor(BaseExecutor): policy: RemoteWorkerSelectionPolicy) -> None: super().__init__() self.workers = workers + self.policy = policy self.worker_count = 0 for worker in self.workers: self.worker_count += worker.count @@ -482,7 +484,6 @@ class RemoteExecutor(BaseExecutor): msg = f"We need somewhere to schedule work; count was {self.worker_count}" logger.critical(msg) raise Exception(msg) - self.policy = policy self.policy.register_worker_pool(self.workers) self.cv = threading.Condition() self._helper_executor = fut.ThreadPoolExecutor( @@ -492,7 +493,7 @@ class RemoteExecutor(BaseExecutor): self.status = RemoteExecutorStatus(self.worker_count) self.total_bundles_submitted = 0 logger.debug( - f'Creating remote processpool with {self.worker_count} remote endpoints.' + f'Creating remote processpool with {self.worker_count} remote worker threads.' ) def is_worker_available(self) -> bool: @@ -586,6 +587,7 @@ class RemoteExecutor(BaseExecutor): bundle.worker = worker machine = bundle.machine = worker.machine username = bundle.username = worker.username + fname = bundle.fname self.status.record_acquire_worker(worker, uuid) logger.debug(f'Running bundle {uuid} on {worker}...') @@ -595,17 +597,17 @@ class RemoteExecutor(BaseExecutor): return self.post_launch_work(bundle) except Exception as e: logger.exception(e) - logger.info(f"Bundle {uuid} seems to have failed?!") + logger.info(f"{uuid}/{fname}: bundle seems to have failed?!") if bundle.failure_count < config.config['executors_max_bundle_failures']: return self.launch(bundle) else: - logger.info(f"Bundle {uuid} is poison, giving up on it.") + logger.info(f"{uuid}/{fname}: bundle is poison, giving up on it.") return None # Send input to machine if it's not local. if hostname not in machine: cmd = f'{RSYNC} {bundle.code_file} {username}@{machine}:{bundle.code_file}' - logger.info(f"Copying work to {worker} via {cmd}") + logger.info(f"{uuid}/{fname}: Copying work to {worker} via {cmd}") run_silently(cmd) # Do it. @@ -615,7 +617,7 @@ class RemoteExecutor(BaseExecutor): f' --code_file {bundle.code_file} --result_file {bundle.result_file}"') p = cmd_in_background(cmd, silent=True) bundle.pid = pid = p.pid - logger.info(f"Running {cmd} in the background as process {pid}") + logger.info(f"{uuid}/{fname}: Start training on {worker} via {cmd} (background pid {pid})") while True: try: @@ -630,7 +632,7 @@ class RemoteExecutor(BaseExecutor): break else: logger.debug( - f"{pid}/{bundle.uuid} has finished its work normally." + f"{uuid}/{fname}: pid {pid} has finished its work normally." ) break @@ -638,10 +640,10 @@ class RemoteExecutor(BaseExecutor): return self.post_launch_work(bundle) except Exception as e: logger.exception(e) - logger.info(f"Bundle {uuid} seems to have failed?!") + logger.info(f"{uuid}: Bundle seems to have failed?!") if bundle.failure_count < config.config['executors_max_bundle_failures']: return self.launch(bundle) - logger.info(f"Bundle {uuid} is poison, giving up on it.") + logger.info(f"{uuid}: Bundle is poison, giving up on it.") return None def post_launch_work(self, bundle: BundleDetails) -> Any: @@ -652,6 +654,8 @@ class RemoteExecutor(BaseExecutor): machine = bundle.machine result_file = bundle.result_file code_file = bundle.code_file + fname = bundle.fname + uuid = bundle.uuid # Whether original or backup, if we finished first we must # fetch the results if the computation happened on a @@ -662,7 +666,7 @@ class RemoteExecutor(BaseExecutor): if bundle.hostname not in bundle.machine: cmd = f'{RSYNC} {username}@{machine}:{result_file} {result_file} 2>/dev/null' logger.info( - f"Fetching results from {username}@{machine} via {cmd}" + f"{uuid}/{fname}: Fetching results from {username}@{machine} via {cmd}" ) try: run_silently(cmd) @@ -686,7 +690,7 @@ class RemoteExecutor(BaseExecutor): # if one of the backups finished first; it still must read the # result from disk. if is_original: - logger.debug(f"Unpickling {result_file}.") + logger.debug(f"{uuid}/{fname}: Unpickling {result_file}.") try: with open(f'{result_file}', 'rb') as rb: serialized = rb.read() @@ -707,7 +711,7 @@ class RemoteExecutor(BaseExecutor): if bundle.backup_bundles is not None: for backup in bundle.backup_bundles: logger.debug( - f'Notifying backup {backup.uuid} that it is cancelled' + f'{uuid}/{fname}: Notifying backup {backup.uuid} that it\'s cancelled' ) backup.is_cancelled.set() @@ -721,7 +725,7 @@ class RemoteExecutor(BaseExecutor): # Tell the original to stop if we finished first. if not was_cancelled: logger.debug( - f'Notifying original {bundle.src_bundle.uuid} that it is cancelled' + f'{uuid}/{fname}: Notifying original {bundle.src_bundle.uuid} that it\'s cancelled' ) bundle.src_bundle.is_cancelled.set() @@ -730,7 +734,7 @@ class RemoteExecutor(BaseExecutor): self.adjust_task_count(-1) return result - def create_original_bundle(self, pickle): + def create_original_bundle(self, pickle, fname: str): from string_utils import generate_uuid uuid = generate_uuid(as_hex=True) code_file = f'/tmp/{uuid}.code.bin' @@ -743,6 +747,7 @@ class RemoteExecutor(BaseExecutor): bundle = BundleDetails( pickled_code = pickle, uuid = uuid, + fname = fname, worker = None, username = None, machine = None, @@ -761,7 +766,7 @@ class RemoteExecutor(BaseExecutor): failure_count = 0, ) self.status.record_bundle_details(bundle) - logger.debug(f'Created original bundle {uuid}') + logger.debug(f'{uuid}/{fname}: Created original bundle') return bundle def create_backup_bundle(self, src_bundle: BundleDetails): @@ -772,6 +777,7 @@ class RemoteExecutor(BaseExecutor): backup_bundle = BundleDetails( pickled_code = src_bundle.pickled_code, uuid = uuid, + fname = src_bundle.fname, worker = None, username = None, machine = None, @@ -791,7 +797,7 @@ class RemoteExecutor(BaseExecutor): ) src_bundle.backup_bundles.append(backup_bundle) self.status.record_bundle_details_already_locked(backup_bundle) - logger.debug(f'Created backup bundle {uuid}') + logger.debug(f'{uuid}/{src_bundle.fname}: Created backup bundle') return backup_bundle def schedule_backup_for_bundle(self, @@ -799,7 +805,7 @@ class RemoteExecutor(BaseExecutor): assert self.status.lock.locked() backup_bundle = self.create_backup_bundle(src_bundle) logger.debug( - f'Scheduling backup bundle {backup_bundle.uuid} for execution' + f'{backup_bundle.uuid}/{backup_bundle.fname}: Scheduling backup for execution...' ) self._helper_executor.submit(self.launch, backup_bundle) @@ -812,7 +818,7 @@ class RemoteExecutor(BaseExecutor): *args, **kwargs) -> fut.Future: pickle = make_cloud_pickle(function, *args, **kwargs) - bundle = self.create_original_bundle(pickle) + bundle = self.create_original_bundle(pickle, function.__name__) self.total_bundles_submitted += 1 return self._helper_executor.submit(self.launch, bundle) @@ -882,7 +888,7 @@ class DefaultExecutors(object): RemoteWorkerRecord( username = 'scott', machine = 'meerkat.cabin', - weight = 6, + weight = 5, count = 2, ), )