def call(self, request, context): command_request = ErCommandRequest.from_proto(request) command_uri = CommandURI(command_request=command_request) service_name = command_uri.get_route() call_result = CommandRouter.get_instance() \ .dispatch(service_name=service_name, args=getattr(command_request, '_args'), kwargs=getattr(command_request, '_kwargs')) response = ErCommandResponse(id=getattr(command_request, '_id'), results=call_result) return response.to_proto()
def put(self, k, v, options: dict = None): if options is None: options = {} k, v = create_serdes(self.__store._store_locator._serdes).serialize(k), \ create_serdes(self.__store._store_locator._serdes).serialize(v) er_pair = ErPair(key=k, value=v) outputs = [] partition_id = self.partitioner(k) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) inputs = [ErPartition(id=partition_id, store_locator=self.__store._store_locator)] output = [ErPartition(id=0, store_locator=self.__store._store_locator)] job_id = generate_job_id(self.__session_id, RollPair.PUT) job = ErJob(id=job_id, name=RollPair.PUT, inputs=[self.__store], outputs=outputs, functors=[ErFunctor(name=RollPair.PUT, body=cloudpickle.dumps(er_pair))]) task = ErTask(id=generate_task_id(job_id, partition_id), name=RollPair.PUT, inputs=inputs, outputs=output, job=job) job_resp = self.__command_client.simple_sync_send( input=task, output_type=ErPair, endpoint=egg._command_endpoint, command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}'), serdes_type=self.__command_serdes ) value = job_resp._value return value
def count(self): total_partitions = self.__store._store_locator._total_partitions job_id = generate_job_id(self.__session_id, tag=RollPair.COUNT) job = ErJob(id=job_id, name=RollPair.COUNT, inputs=[self.ctx.populate_processor(self.__store)]) args = list() for i in range(total_partitions): partition_input = job._inputs[0]._partitions[i] task = ErTask(id=generate_task_id(job_id, i), name=job._name, inputs=[partition_input], job=job) args.append(([task], partition_input._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI( f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) done = wait(futures, return_when=FIRST_EXCEPTION).done result = 0 for future in done: pair = future.result()[0] result += self.functor_serdes.deserialize(pair._value) return result
def map_values(self, func, output=None, options: dict = None): if options is None: options = {} functor = ErFunctor(name=RollPair.MAP_VALUES, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) outputs = [] if output: outputs.append(output) # todo:1: options issues. refer to line 77 final_options = {} final_options.update(self.__store._options) final_options.update(options) job = ErJob(id=generate_job_id(self.__session_id, RollPair.MAP_VALUES), name=RollPair.MAP_VALUES, inputs=[self.__store], outputs=outputs, functors=[functor], options=final_options) job_result = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=self.__command_serdes) er_store = job_result._outputs[0] return RollPair(er_store, self.ctx)
def subtract_by_key(self, other, output=None, options: dict = None): if options is None: options = {} functor = ErFunctor(name=RollPair.SUBTRACT_BY_KEY, serdes=SerdesTypes.CLOUD_PICKLE) outputs = [] if output: outputs.append(output) job = ErJob(id=generate_job_id(self.__session_id, RollPair.SUBTRACT_BY_KEY), name=RollPair.SUBTRACT_BY_KEY, inputs=self.__repartition_with(other), outputs=outputs, functors=[functor]) job_result = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=self.__command_serdes) er_store = job_result._outputs[0] return RollPair(er_store, self.ctx)
def flat_map(self, func, output=None, options: dict = None): if options is None: options = {} functor = ErFunctor(name=RollPair.FLAT_MAP, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) outputs = [] if output: outputs.append(output) job = ErJob(id=generate_job_id(self.__session_id, RollPair.FLAT_MAP), name=RollPair.FLAT_MAP, inputs=[self.__store], outputs=outputs, functors=[functor]) job_result = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=self.__command_serdes) er_store = job_result._outputs[0] return RollPair(er_store, self.ctx)
def sample(self, fraction, seed=None, output=None, options: dict = None): if options is None: options = {} er_fraction = ErFunctor(name=RollPair.REDUCE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(fraction)) er_seed = ErFunctor(name=RollPair.REDUCE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(seed)) outputs = [] if output: outputs.append(output) job = ErJob(id=generate_job_id(self.__session_id, RollPair.SAMPLE), name=RollPair.SAMPLE, inputs=[self.__store], outputs=outputs, functors=[er_fraction, er_seed]) job_result = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=self.__command_serdes) er_store = job_result._outputs[0] L.info(er_store) return RollPair(er_store, self.ctx)
def send_command(tasks, remote_cmd_endpoint): cmd_client = CommandClient() return cmd_client.sync_send( inputs=tasks, output_types=[ErTask], endpoint=remote_cmd_endpoint, command_uri=CommandURI(f'v1/egg-pair/runTask'))
def destroy(self): if len(self.ctx.get_session()._cluster_manager_client.get_store( self.get_store())._partitions) == 0: L.info(f"store:{self.get_store()} has been destroyed before") raise ValueError( f"store:{self.get_store()} has been destroyed before") total_partitions = self.__store._store_locator._total_partitions job = ErJob(id=generate_job_id(self.__session_id, RollPair.DESTROY), name=RollPair.DESTROY, inputs=[self.__store], outputs=[self.__store], functors=[]) job_resp = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=self.__command_serdes) self.ctx.get_session()._cluster_manager_client.delete_store( self.__store) L.info(f'{RollPair.DESTROY}: {self.__store}') self.destroyed = True
class NodeManagerCommands(object): prefix = 'v1/node-manager/processor' get_or_create_processor_batch = 'getOrCreateProcessorBatch' get_or_create_processor_batch_service_name = _to_service_name( prefix, get_or_create_processor_batch) GET_OR_CREATE_PROCESSOR_BATCH = CommandURI( get_or_create_processor_batch_service_name) get_or_create_servicer = 'getOrCreateServicer' get_or_create_servicer_service_name = _to_service_name( prefix, get_or_create_servicer) GET_OR_CREATE_SERVICER = CommandURI(get_or_create_servicer_service_name) heartbeat = 'heartbeat' heartbeat_service_name = _to_service_name(prefix, heartbeat) HEARTBEAT = CommandURI(heartbeat_service_name)
class SessionCommands(object): prefix = 'v1/cluster-manager/session' get_or_create_session = 'getOrCreateSession' get_or_create_session_service_name = _to_service_name( prefix, get_or_create_session) GET_OR_CREATE_SESSION = CommandURI(get_or_create_session_service_name) register_session = 'registerSession' register_session_service_name = _to_service_name(prefix, register_session) REGISTER_SESSION = CommandURI(register_session_service_name) get_session_server_nodes = 'getSessionServerNodes' get_session_server_nodes_service_name = _to_service_name( prefix, get_session_server_nodes) GET_SESSION_SERVER_NODES = CommandURI( get_session_server_nodes_service_name) get_session_rolls = "getSessionRolls" get_session_rolls_service_name = _to_service_name(prefix, get_session_rolls) GET_SESSION_ROLLS = CommandURI(get_session_rolls_service_name) get_session_eggs = "getSessionEggs" get_session_eggs_service_name = _to_service_name(prefix, get_session_eggs) GET_SESSION_EGGS = CommandURI(get_session_eggs_service_name) heartbeat = 'heartbeat' heartbeat_service_name = _to_service_name(prefix, heartbeat) HEARTBEAT = CommandURI(heartbeat_service_name) stop_session = 'stopSession' stop_session_service_name = _to_service_name(prefix, stop_session) STOP_SESSION = CommandURI(stop_session_service_name) kill_session = 'killSession' kill_session_service_name = _to_service_name(prefix, kill_session) KILL_SESSION = CommandURI(kill_session_service_name) kill_all_sessions = "killAllSessions" kill_all_sessions_service_name = _to_service_name(prefix, kill_all_sessions) KILL_ALL_SESSIONS = CommandURI(kill_all_sessions_service_name)
def aggregate(self, zero_value, seq_op, comb_op, output=None, options: dict = None): total_partitions = self.__store._store_locator._total_partitions job_id = generate_job_id(self.__session_id, tag=RollPair.AGGREGATE) serialized_zero_value = ErFunctor(name=RollPair.AGGREGATE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(zero_value)) serialized_seq_op = ErFunctor(name=RollPair.AGGREGATE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(seq_op)) job = ErJob(id=job_id, name=RollPair.AGGREGATE, inputs=[self.ctx.populate_processor(self.__store)], functors=[serialized_zero_value, serialized_seq_op]) args = list() for i in range(total_partitions): partition_input = job._inputs[0]._partitions[i] task = ErTask(id=generate_task_id(job_id, i), name=job._name, inputs=[partition_input], job=job) args.append(([task], partition_input._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI( f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) done = wait(futures, return_when=FIRST_EXCEPTION).done result = None first = True for future in done: pair = future.result()[0] seq_op_result = self.functor_serdes.deserialize(pair._value) if not first: result = comb_op(result, seq_op_result) else: result = seq_op_result first = False return result
def send_command(): job = ErJob(id=job_id, name=RollPair.PUT_ALL, inputs=[self.__store], outputs=[self.__store], functors=[]) result = self.__command_client.simple_sync_send( input=job, output_type=ErJob, endpoint=self.ctx.get_roll()._command_endpoint, command_uri=CommandURI( f'{RollPair.ROLL_PAIR_URI_PREFIX}/{RollPair.RUN_JOB}'), serdes_type=SerdesTypes.PROTOBUF) return result
def with_stores(self, func, others=None, options: dict = None): if options is None: options = {} tag = "withStores" if others is None: others = [] total_partitions = self.get_partitions() for other in others: if other.get_partitions() != total_partitions: raise ValueError( f"diff partitions: expected:{total_partitions}, actual:{other.get_partitions()}" ) job_id = generate_job_id(self.__session_id, tag=tag) job = ErJob(id=job_id, name=tag, inputs=[ self.ctx.populate_processor(rp.get_store()) for rp in [self] + others ], functors=[ ErFunctor(name=tag, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) ], options=options) args = list() for i in range(total_partitions): partition_self = job._inputs[0]._partitions[i] task = ErTask( id=generate_task_id(job_id, i), name=job._name, inputs=[store._partitions[i] for store in job._inputs], job=job) args.append(([task], partition_self._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI( f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) result = list() for future in futures: ret_pair = future.result()[0] result.append((self.functor_serdes.deserialize(ret_pair._key), self.functor_serdes.deserialize(ret_pair._value))) return result
def get(self, k, options: dict = None): if options is None: options = {} L.debug(f"get k: {k}") k = create_serdes(self.__store._store_locator._serdes).serialize(k) er_pair = ErPair(key=k, value=None) outputs = [] value = None partition_id = self.partitioner(k) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) L.info( f"partitions count: {self.__store._store_locator._total_partitions}, target partition: {partition_id}, endpoint: {egg._command_endpoint}" ) inputs = [ ErPartition(id=partition_id, store_locator=self.__store._store_locator) ] output = [ ErPartition(id=partition_id, store_locator=self.__store._store_locator) ] job_id = generate_job_id(self.__session_id, RollPair.GET) job = ErJob(id=job_id, name=RollPair.GET, inputs=[self.__store], outputs=outputs, functors=[ErFunctor(body=cloudpickle.dumps(er_pair))]) task = ErTask(id=generate_task_id(job_id, partition_id), name=RollPair.GET, inputs=inputs, outputs=output, job=job) job_resp = self.__command_client.simple_sync_send( input=task, output_type=ErPair, endpoint=egg._command_endpoint, command_uri=CommandURI( f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}'), serdes_type=self.__command_serdes) return self.value_serdes.deserialize( job_resp._value) if job_resp._value != b'' else None
class MetadataCommands(object): prefix = 'v1/cluster-manager/metadata' get_server_node = 'getServerNode' get_server_node_service_name = _to_service_name(prefix, get_server_node) GET_SERVER_NODE = CommandURI(get_server_node_service_name) get_server_nodes = 'getServerNodes' get_server_nodes_service_name = _to_service_name(prefix, get_server_nodes) GET_SERVER_NODES = CommandURI(get_server_nodes_service_name) get_or_create_server_node = 'getOrCreateServerNode' get_or_create_server_node_service_name = _to_service_name( prefix, get_or_create_server_node) GET_OR_CREATE_SERVER_NODE = CommandURI( get_or_create_server_node_service_name) create_or_update_server_node = 'createOrUpdateServerNode' create_or_update_server_node_service_name = _to_service_name( prefix, create_or_update_server_node) CREATE_OR_UPDATE_SERVER_NODE = CommandURI( create_or_update_server_node_service_name) get_store = 'getStore' get_store_service_name = _to_service_name(prefix, get_store) GET_STORE = CommandURI(get_store_service_name) get_or_create_store = 'getOrCreateStore' get_or_create_store_service_name = _to_service_name( prefix, get_or_create_store) GET_OR_CREATE_STORE = CommandURI(get_or_create_store_service_name) delete_store = 'deleteStore' delete_store_service_name = _to_service_name(prefix, delete_store) DELETE_STORE = CommandURI(delete_store_service_name) get_store_from_namespace = 'getStoreFromNamespace' get_store_from_namespace_service_name = _to_service_name( prefix, get_store_from_namespace) GET_STORE_FROM_NAMESPACE = CommandURI( get_store_from_namespace_service_name)
def delete(self, k, options: dict = None): if options is None: options = {} key = create_serdes(self.__store._store_locator._serdes).serialize(k) er_pair = ErPair(key=key, value=None) outputs = [] value = None partition_id = self.partitioner(key) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) L.info(egg._command_endpoint) L.info(f"count: {self.__store._store_locator._total_partitions}") inputs = [ ErPartition(id=partition_id, store_locator=self.__store._store_locator) ] output = [ ErPartition(id=partition_id, store_locator=self.__store._store_locator) ] job_id = generate_job_id(self.__session_id, RollPair.DELETE) job = ErJob(id=job_id, name=RollPair.DELETE, inputs=[self.__store], outputs=outputs, functors=[ErFunctor(body=cloudpickle.dumps(er_pair))]) task = ErTask(id=generate_task_id(job_id, partition_id), name=RollPair.DELETE, inputs=inputs, outputs=output, job=job) L.info("start send req") job_resp = self.__command_client.simple_sync_send( input=task, output_type=ErPair, endpoint=egg._command_endpoint, command_uri=CommandURI( f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}'), serdes_type=self.__command_serdes)
class RollPair(object): ROLL_PAIR_URI_PREFIX = 'v1/roll-pair' EGG_PAIR_URI_PREFIX = 'v1/egg-pair' RUN_JOB = 'runJob' RUN_TASK = 'runTask' AGGREGATE = 'aggregate' COLLAPSE_PARTITIONS = 'collapsePartitions' CLEANUP = 'cleanup' COUNT = 'count' DELETE = "delete" DESTROY = "destroy" FILTER = 'filter' FLAT_MAP = 'flatMap' GET = "get" GET_ALL = "getAll" GLOM = 'glom' JOIN = 'join' MAP = 'map' MAP_PARTITIONS = 'mapPartitions' MAP_VALUES = 'mapValues' PUT = "put" PUT_ALL = "putAll" REDUCE = 'reduce' SAMPLE = 'sample' SUBTRACT_BY_KEY = 'subtractByKey' UNION = 'union' RUN_TASK_URI = CommandURI(f'{EGG_PAIR_URI_PREFIX}/{RUN_TASK}') SERIALIZED_NONE = cloudpickle.dumps(None) def __setstate__(self, state): self.gc_enable = None pass def __getstate__(self): pass def __init__(self, er_store: ErStore, rp_ctx: RollPairContext): if not rp_ctx: raise ValueError('rp_ctx cannot be None') self.__store = er_store self.ctx = rp_ctx self.__command_serdes = SerdesTypes.PROTOBUF #self.__roll_pair_master = self.ctx.get_roll() self.__command_client = CommandClient() self.functor_serdes =create_serdes(SerdesTypes.CLOUD_PICKLE) self.value_serdes = self.get_store_serdes() self.key_serdes = self.get_store_serdes() self.partitioner = partitioner(hash_code, self.__store._store_locator._total_partitions) self.egg_router = default_egg_router self.__session_id = self.ctx.session_id self.gc_enable = rp_ctx.rpc_gc_enable self.gc_recorder = rp_ctx.gc_recorder self.gc_recorder.record(er_store) self.destroyed = False def __del__(self): if "EGGROLL_GC_DISABLE" in os.environ and os.environ["EGGROLL_GC_DISABLE"] == '1': L.trace("global RollPair gc is disable") return if not hasattr(self, 'gc_enable') \ or not hasattr(self, 'ctx'): return if not self.gc_enable: L.debug('GC not enabled session={}'.format(self.__session_id)) return if self.get_store_type() != StoreTypes.ROLLPAIR_IN_MEMORY: return if self.destroyed: return if self.ctx.get_session().is_stopped(): L.trace('session={} has already been stopped'.format(self.__session_id)) return self.ctx.gc_recorder.decrease_ref_count(self.__store) def __repr__(self): return f'<RollPair(_store={self.__store}) at {hex(id(self))}>' @staticmethod def __check_partition(input_partitions, output_partitions, shuffle=False): if not shuffle: return if input_partitions != output_partitions: raise ValueError(f"input partitions:{input_partitions}, output partitions:{output_partitions}," f"must be the same!") def __repartition_with(self, other): self_partition = self.get_partitions() other_partition = other.get_partitions() should_shuffle = False if len(self.__store._partitions) != len(other.__store._partitions): should_shuffle = True else: for i in range(len(self.__store._partitions)): if self.__store._partitions[i]._processor._id != other.__store._partitions[i]._processor._id: should_shuffle = True if other_partition != self_partition or should_shuffle: self_name = self.get_name() self_count = self.count() other_name = other.get_name() other_count = other.count() L.debug(f"repartition start: self rp={self_name} partitions={self_partition}, " f"other={other_name}: partitions={other_partition}, repartitioning") if self_count < other_count: shuffle_rp = self shuffle_rp_count = self_count shuffle_rp_name = self_name shuffle_total_partitions = self_partition shuffle_rp_partitions = self.__store._partitions not_shuffle_rp = other not_shuffle_rp_count = other_count not_shuffle_rp_name = other_name not_shuffle_total_partitions = other_partition not_shuffle_rp_partitions = other.__store._partitions else: not_shuffle_rp = self not_shuffle_rp_count = self_count not_shuffle_rp_name = self_name not_shuffle_total_partitions = self_partition not_shuffle_rp_partitions = self.__store._partitions shuffle_rp = other shuffle_rp_count = other_count shuffle_rp_name = other_name shuffle_total_partitions = other_partition shuffle_rp_partitions = other.__store._partitions L.trace(f"repartition selection: rp={shuffle_rp_name} count={shuffle_rp_count}, " f"rp={not_shuffle_rp_name} count={not_shuffle_rp_count}. " f"repartitioning {shuffle_rp_name}") store = ErStore(store_locator=ErStoreLocator(store_type=shuffle_rp.get_store_type(), namespace=shuffle_rp.get_namespace(), name=str(uuid.uuid1()), total_partitions=not_shuffle_total_partitions), partitions=not_shuffle_rp_partitions) res_rp = shuffle_rp.map(lambda k, v: (k, v), output=store) res_rp.disable_gc() if L.isEnabledFor(logging.DEBUG): L.debug(f"repartition end: rp to shuffle={shuffle_rp_name}, " f"count={shuffle_rp_count}, partitions={shuffle_total_partitions}; " f"rp NOT shuffled={not_shuffle_rp_name}, " f"count={not_shuffle_rp_count}, partitions={not_shuffle_total_partitions}' " f"res rp={res_rp.get_name()}, " f"count={res_rp.count()}, partitions={res_rp.get_partitions()}") store_shuffle = res_rp.get_store() return [store_shuffle, other.get_store()] if self_count < other_count \ else [self.get_store(), store_shuffle] else: return [self.__store, other.__store] def enable_gc(self): self.gc_enable = True def disable_gc(self): self.gc_enable = False def get_store_serdes(self): return create_serdes(self.__store._store_locator._serdes) def get_partitions(self): return self.__store._store_locator._total_partitions def get_name(self): return self.__store._store_locator._name def get_namespace(self): return self.__store._store_locator._namespace def get_store(self): return self.__store def get_store_type(self): return self.__store._store_locator._store_type def kv_to_bytes(self, **kwargs): use_serialize = kwargs.get("use_serialize", True) # can not use is None if "k" in kwargs and "v" in kwargs: k, v = kwargs["k"], kwargs["v"] return (self.value_serdes.serialize(k), self.value_serdes.serialize(v)) if use_serialize \ else (string_to_bytes(k), string_to_bytes(v)) elif "k" in kwargs: k = kwargs["k"] return self.value_serdes.serialize(k) if use_serialize else string_to_bytes(k) elif "v" in kwargs: v = kwargs["v"] return self.value_serdes.serialize(v) if use_serialize else string_to_bytes(v) def _run_job(self, job: ErJob, output_types: list = None, command_uri: CommandURI = RUN_TASK_URI, create_output_if_missing: bool = True): from eggroll.core.utils import _map_and_listify start = time.time() L.debug(f"[RUNJOB] calling: job_id={job._id}, name={job._name}, inputs={_map_and_listify(lambda i: f'namespace={i._store_locator._namespace}, name={i._store_locator._name}, store_type={i._store_locator._store_type}, total_partitions={i._store_locator._total_partitions}', job._inputs)}") futures = self.ctx.get_session().submit_job( job=job, output_types=output_types, command_uri=command_uri, create_output_if_missing=create_output_if_missing) results = list() for future in futures: results.append(future.result()) elapsed = time.time() - start L.debug(f"[RUNJOB] called (elapsed={elapsed}): job_id={job._id}, name={job._name}, inputs={_map_and_listify(lambda i: f'namespace={i._store_locator._namespace}, name={i._store_locator._name}, store_type={i._store_locator._store_type}, total_partitions={i._store_locator._total_partitions}', job._inputs)}") return results def __get_output_from_result(self, results): return results[0][0]._job._outputs[0] """ storage api """ @_method_profile_logger def get(self, k, options: dict = None): if options is None: options = {} k = create_serdes(self.__store._store_locator._serdes).serialize(k) er_pair = ErPair(key=k, value=None) partition_id = self.partitioner(k) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) inputs = [ErPartition(id=partition_id, store_locator=self.__store._store_locator)] outputs = [ErPartition(id=partition_id, store_locator=self.__store._store_locator)] job_id = generate_job_id(self.__session_id, RollPair.GET) job = ErJob(id=job_id, name=RollPair.GET, inputs=[self.__store], outputs=[self.__store], functors=[ErFunctor(name=RollPair.GET, body=cloudpickle.dumps(er_pair))]) task = ErTask(id=generate_task_id(job_id, partition_id), name=RollPair.GET, inputs=inputs, outputs=outputs, job=job) job_resp = self.__command_client.simple_sync_send( input=task, output_type=ErPair, endpoint=egg._command_endpoint, command_uri=self.RUN_TASK_URI, serdes_type=self.__command_serdes ) return self.value_serdes.deserialize(job_resp._value) if job_resp._value != b'' else None @_method_profile_logger def put(self, k, v, options: dict = None): if options is None: options = {} k, v = create_serdes(self.__store._store_locator._serdes).serialize(k), \ create_serdes(self.__store._store_locator._serdes).serialize(v) er_pair = ErPair(key=k, value=v) outputs = [] partition_id = self.partitioner(k) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) inputs = [ErPartition(id=partition_id, store_locator=self.__store._store_locator)] output = [ErPartition(id=0, store_locator=self.__store._store_locator)] job_id = generate_job_id(self.__session_id, RollPair.PUT) job = ErJob(id=job_id, name=RollPair.PUT, inputs=[self.__store], outputs=outputs, functors=[ErFunctor(name=RollPair.PUT, body=cloudpickle.dumps(er_pair))]) task = ErTask(id=generate_task_id(job_id, partition_id), name=RollPair.PUT, inputs=inputs, outputs=output, job=job) job_resp = self.__command_client.simple_sync_send( input=task, output_type=ErPair, endpoint=egg._command_endpoint, command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}'), serdes_type=self.__command_serdes ) value = job_resp._value return value @_method_profile_logger def get_all(self, limit=None, options: dict = None): if options is None: options = {} if limit is not None and not isinstance(limit, int) and limit <= 0: raise ValueError(f"limit:{limit} must be positive int") job_id = generate_job_id(self.__session_id, RollPair.GET_ALL) er_pair = ErPair(key=create_serdes(self.__store._store_locator._serdes) .serialize(limit) if limit is not None else None, value=None) def send_command(): job = ErJob(id=job_id, name=RollPair.GET_ALL, inputs=[self.__store], outputs=[self.__store], functors=[ErFunctor(name=RollPair.GET_ALL, body=cloudpickle.dumps(er_pair))]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return er_store send_command() populated_store = self.ctx.populate_processor(self.__store) transfer_pair = TransferPair(transfer_id=job_id) done_cnt = 0 for k, v in transfer_pair.gather(populated_store): done_cnt += 1 yield self.key_serdes.deserialize(k), self.value_serdes.deserialize(v) L.trace(f"get_all: namespace={self.get_namespace()} name={self.get_name()}, count={done_cnt}") @_method_profile_logger def put_all(self, items, output=None, options: dict = None): if options is None: options = {} include_key = options.get("include_key", True) job_id = generate_job_id(self.__session_id, RollPair.PUT_ALL) # TODO:1: consider multiprocessing scenario. parallel size should be sent to egg_pair to set write signal count def send_command(): job = ErJob(id=job_id, name=RollPair.PUT_ALL, inputs=[self.__store], outputs=[self.__store], functors=[]) task_results = self._run_job(job) return self.__get_output_from_result(task_results) th = Thread(target=send_command, name=f'roll_pair-send_command-{job_id}') th.start() populated_store = self.ctx.populate_processor(self.__store) shuffler = TransferPair(job_id) fifo_broker = FifoBroker() bb = BatchBroker(fifo_broker) scatter_future = shuffler.scatter(fifo_broker, self.partitioner, populated_store) key_serdes = self.key_serdes value_serdes = self.value_serdes try: if include_key: for k, v in items: bb.put(item=(key_serdes.serialize(k), value_serdes.serialize(v))) else: k = 0 for v in items: bb.put(item=(key_serdes.serialize(k), value_serdes.serialize(v))) k += 1 finally: bb.signal_write_finish() scatter_results = scatter_future.result() th.join() return RollPair(populated_store, self.ctx) @_method_profile_logger def count(self): job_id = generate_job_id(self.__session_id, tag=RollPair.COUNT) job = ErJob(id=job_id, name=RollPair.COUNT, inputs=[self.__store]) task_results = self._run_job(job=job, output_types=[ErPair], create_output_if_missing=False) result = 0 for task_result in task_results: pair = task_result[0] result += self.functor_serdes.deserialize(pair._value) return result # todo:1: move to command channel to utilize batch command @_method_profile_logger def destroy(self, options: dict = None): if len(self.ctx.get_session()._cluster_manager_client.get_store(self.get_store())._partitions) == 0: L.exception(f"store:{self.get_store()} has been destroyed before") raise ValueError(f"store:{self.get_store()} has been destroyed before") if options is None: options = {} job = ErJob(id=generate_job_id(self.__session_id, RollPair.DESTROY), name=RollPair.DESTROY, inputs=[self.__store], outputs=[self.__store], functors=[], options=options) task_results = self._run_job(job=job, create_output_if_missing=False) self.ctx.get_session()._cluster_manager_client.delete_store(self.__store) L.debug(f'{RollPair.DESTROY}={self.__store}') self.destroyed = True @_method_profile_logger def delete(self, k, options: dict = None): if options is None: options = {} key = create_serdes(self.__store._store_locator._serdes).serialize(k) er_pair = ErPair(key=key, value=None) value = None partition_id = self.partitioner(key) egg = self.ctx.route_to_egg(self.__store._partitions[partition_id]) job_id = generate_job_id(self.__session_id, RollPair.DELETE) job = ErJob(id=job_id, name=RollPair.DELETE, inputs=[self.__store], outputs=[], functors=[ErFunctor(name=RollPair.DELETE, body=cloudpickle.dumps(er_pair))]) task_results = self._run_job(job=job, create_output_if_missing=False) @_method_profile_logger def take(self, n: int, options: dict = None): if options is None: options = {} if n <= 0: n = 1 keys_only = options.get("keys_only", False) ret = [] count = 0 for item in self.get_all(limit=n): if keys_only: if item: ret.append(item[0]) else: ret.append(None) else: ret.append(item) count += 1 if count == n: break return ret @_method_profile_logger def first(self, options: dict = None): if options is None: options = {} resp = self.take(1, options=options) if resp: return resp[0] else: return None @_method_profile_logger def save_as(self, name=None, namespace=None, partition=None, options: dict = None): if partition is not None and partition <= 0: raise ValueError('partition cannot <= 0') if not namespace: namespace = self.get_namespace() if not name: if self.get_namespace() == namespace: forked_store_locator = self.get_store()._store_locator.fork() name = forked_store_locator._name else: name = self.get_name() if not partition: partition = self.get_partitions() if options is None: options = {} store_type = options.get('store_type', self.ctx.default_store_type) refresh_nodes = options.get('refresh_nodes') saved_as_store = ErStore(store_locator=ErStoreLocator( store_type=store_type, namespace=namespace, name=name, total_partitions=partition)) if partition == self.get_partitions() and not refresh_nodes: return self.map_values(lambda v: v, output=saved_as_store, options=options) else: return self.map(lambda k, v: (k, v), output=saved_as_store, options=options) """ computing api """ @_method_profile_logger def map_values(self, func, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.MAP_VALUES, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) # todo:1: options issues. refer to line 77 final_options = {} final_options.update(self.__store._options) final_options.update(options) job = ErJob(id=generate_job_id(self.__session_id, RollPair.MAP_VALUES), name=RollPair.MAP_VALUES, inputs=[self.__store], outputs=outputs, functors=[functor], options=final_options) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def map(self, func, output=None, options: dict = None): if options is None: options = {} functor = ErFunctor(name=RollPair.MAP, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.MAP), name=RollPair.MAP, inputs=[self.__store], outputs=[output], functors=[functor], options=options) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def map_partitions(self, func, reduce_op=None, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) shuffle = options.get('shuffle', True) if not shuffle and reduce_op: raise ValueError(f"shuffle cannot be False when reduce is needed!") functor = ErFunctor(name=RollPair.MAP_PARTITIONS, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) reduce_functor = ErFunctor(name=RollPair.MAP_PARTITIONS, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(reduce_op)) need_shuffle = ErFunctor(name=RollPair.MAP_PARTITIONS, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(shuffle)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.MAP_PARTITIONS), name=RollPair.MAP_PARTITIONS, inputs=[self.__store], outputs=outputs, functors=[functor, reduce_functor, need_shuffle]) task_future = self._run_job(job=job) er_store = self.__get_output_from_result(task_future) return RollPair(er_store, self.ctx) @_method_profile_logger def collapse_partitions(self, func, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.COLLAPSE_PARTITIONS, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.COLLAPSE_PARTITIONS), name=RollPair.COLLAPSE_PARTITIONS, inputs=[self.__store], outputs=outputs, functors=[functor]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def flat_map(self, func, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) shuffle = options.get('shuffle', True) functor = ErFunctor(name=RollPair.FLAT_MAP, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) need_shuffle = ErFunctor(name=RollPair.FLAT_MAP, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(shuffle)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.FLAT_MAP), name=RollPair.FLAT_MAP, inputs=[self.__store], outputs=outputs, functors=[functor, need_shuffle]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def reduce(self, func, output=None, options: dict = None): total_partitions = self.__store._store_locator._total_partitions job_id = generate_job_id(self.__session_id, tag=RollPair.REDUCE) serialized_func = ErFunctor(name=RollPair.REDUCE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) job = ErJob(id=job_id, name=RollPair.REDUCE, inputs=[self.ctx.populate_processor(self.__store)], functors=[serialized_func]) args = list() for i in range(total_partitions): partition_input = job._inputs[0]._partitions[i] task = ErTask(id=generate_task_id(job_id, i), name=job._name, inputs=[partition_input], job=job) args.append(([task], partition_input._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) done = wait(futures, return_when=FIRST_EXCEPTION).done result = None first = True for future in done: pair = future.result()[0] seq_op_result = self.functor_serdes.deserialize(pair._value) if seq_op_result is not None: if not first: result = func(result, seq_op_result) else: result = seq_op_result first = False return result @_method_profile_logger def aggregate(self, zero_value, seq_op, comb_op, output=None, options: dict = None): total_partitions = self.__store._store_locator._total_partitions job_id = generate_job_id(self.__session_id, tag=RollPair.AGGREGATE) serialized_zero_value = ErFunctor(name=RollPair.AGGREGATE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(zero_value)) serialized_seq_op = ErFunctor(name=RollPair.AGGREGATE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(seq_op)) job = ErJob(id=job_id, name=RollPair.AGGREGATE, inputs=[self.ctx.populate_processor(self.__store)], functors=[serialized_zero_value, serialized_seq_op]) args = list() for i in range(total_partitions): partition_input = job._inputs[0]._partitions[i] task = ErTask(id=generate_task_id(job_id, i), name=job._name, inputs=[partition_input], job=job) args.append(([task], partition_input._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) done = wait(futures, return_when=FIRST_EXCEPTION).done result = None first = True for future in done: pair = future.result()[0] seq_op_result = self.functor_serdes.deserialize(pair._value) if not first: result = comb_op(result, seq_op_result) else: result = seq_op_result first = False return result @_method_profile_logger def glom(self, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.GLOM, serdes=SerdesTypes.CLOUD_PICKLE) job = ErJob(id=generate_job_id(self.__session_id, RollPair.GLOM), name=RollPair.GLOM, inputs=[self.__store], outputs=outputs, functors=[functor]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def sample(self, fraction, seed=None, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) er_fraction = ErFunctor(name=RollPair.REDUCE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(fraction)) er_seed = ErFunctor(name=RollPair.REDUCE, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(seed)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.SAMPLE), name=RollPair.SAMPLE, inputs=[self.__store], outputs=outputs, functors=[er_fraction, er_seed]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def filter(self, func, output=None, options: dict = None): if options is None: options = {} outputs = [] if output: RollPair.__check_partition(self.get_partitions(), output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.FILTER, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.FILTER), name=RollPair.FILTER, inputs=[self.__store], outputs=outputs, functors=[functor]) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def subtract_by_key(self, other, output=None, options: dict = None): if options is None: options = {} inputs = self.__repartition_with(other) outputs = [] if output: RollPair.__check_partition(inputs[0]._store_locator._total_partitions, output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.SUBTRACT_BY_KEY, serdes=SerdesTypes.CLOUD_PICKLE) job = ErJob(id=generate_job_id(self.__session_id, RollPair.SUBTRACT_BY_KEY), name=RollPair.SUBTRACT_BY_KEY, inputs=inputs, outputs=outputs, functors=[functor]) task_future = self._run_job(job=job) er_store = self.__get_output_from_result(task_future) return RollPair(er_store, self.ctx) @_method_profile_logger def union(self, other, func=lambda v1, v2: v1, output=None, options: dict = None): if options is None: options = {} inputs = self.__repartition_with(other) outputs = [] if output: RollPair.__check_partition(inputs[0]._store_locator._total_partitions, output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.UNION, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) job = ErJob(id=generate_job_id(self.__session_id, RollPair.UNION), name=RollPair.UNION, inputs=inputs, outputs=outputs, functors=[functor]) task_future = self._run_job(job=job) er_store = self.__get_output_from_result(task_future) return RollPair(er_store, self.ctx) @_method_profile_logger def join(self, other, func, output=None, options: dict = None): if options is None: options = {} inputs = self.__repartition_with(other) outputs = [] if output: RollPair.__check_partition(inputs[0]._store_locator._total_partitions, output._store_locator._total_partitions) outputs.append(output) functor = ErFunctor(name=RollPair.JOIN, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func)) final_options = {} final_options.update(self.__store._options) final_options.update(options) job = ErJob(id=generate_job_id(self.__session_id, RollPair.JOIN), name=RollPair.JOIN, inputs=inputs, outputs=outputs, functors=[functor], options=final_options) task_results = self._run_job(job=job) er_store = self.__get_output_from_result(task_results) return RollPair(er_store, self.ctx) @_method_profile_logger def with_stores(self, func, others=None, options: dict = None): if options is None: options = {} tag = "withStores" if others is None: others = [] total_partitions = self.get_partitions() for other in others: if other.get_partitions() != total_partitions: raise ValueError(f"diff partitions: expected:{total_partitions}, actual:{other.get_partitions()}") job_id = generate_job_id(self.__session_id, tag=tag) job = ErJob(id=job_id, name=tag, inputs=[self.ctx.populate_processor(rp.get_store()) for rp in [self] + others], functors=[ErFunctor(name=tag, serdes=SerdesTypes.CLOUD_PICKLE, body=cloudpickle.dumps(func))], options=options) args = list() for i in range(total_partitions): partition_self = job._inputs[0]._partitions[i] task = ErTask(id=generate_task_id(job_id, i), name=job._name, inputs=[store._partitions[i] for store in job._inputs], job=job) args.append(([task], partition_self._processor._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErPair], command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) result = list() for future in futures: ret_pair = future.result()[0] result.append((self.functor_serdes.deserialize(ret_pair._key), self.functor_serdes.deserialize(ret_pair._value))) return result
def cleanup(self, name, namespace, options: dict = None): if not namespace: raise ValueError('namespace cannot be blank') L.debug(f'cleaning up namespace={namespace}, name={name}') if options is None: options = {} total_partitions = options.get('total_partitions', 1) partitioner = options.get('partitioner', PartitionerTypes.BYTESTRING_HASH) store_serdes = options.get('serdes', self.default_store_serdes) if name == '*': store_type = options.get('store_type', '*') L.debug(f'cleaning up whole store_type={store_type}, namespace={namespace}, name={name}') er_store = ErStore(store_locator=ErStoreLocator(namespace=namespace, name=name, store_type=store_type)) job_id = generate_job_id(namespace, tag=RollPair.CLEANUP) job = ErJob(id=job_id, name=RollPair.DESTROY, inputs=[er_store], options=options) args = list() cleanup_partitions = [ErPartition(id=-1, store_locator=er_store._store_locator)] for server_node, eggs in self.__session._eggs.items(): egg = eggs[0] task = ErTask(id=generate_task_id(job_id, egg._command_endpoint._host), name=job._name, inputs=cleanup_partitions, job=job) args.append(([task], egg._command_endpoint)) futures = self.__command_client.async_call( args=args, output_types=[ErTask], command_uri=CommandURI(f'{RollPair.EGG_PAIR_URI_PREFIX}/{RollPair.RUN_TASK}')) for future in futures: result = future.result() self.get_session()._cluster_manager_client.delete_store(er_store) else: # todo:1: add combine options to pass it through store_options = self.__session.get_all_options() store_options.update(options) final_options = store_options.copy() store = ErStore( store_locator=ErStoreLocator( store_type=StoreTypes.ROLLPAIR_LMDB, namespace=namespace, name=name, total_partitions=total_partitions, partitioner=partitioner, serdes=store_serdes), options=final_options) task_results = self.__session._cluster_manager_client.get_store_from_namespace(store) L.trace('res={}'.format(task_results._stores)) if task_results._stores is not None: L.trace("item count={}".format(len(task_results._stores))) for item in task_results._stores: L.trace("item namespace={} name={}".format(item._store_locator._namespace, item._store_locator._name)) rp = RollPair(er_store=item, rp_ctx=self) rp.destroy()