role_maker.py

#   Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Defination of Role Makers."""

from multiprocessing import Process, Manager
import paddle.fluid as fluid
import os
import time

__all__ = [
    'Role', 'RoleMakerBase', 'MPISymetricRoleMaker', 'UserDefinedRoleMaker',
    'UserDefinedCollectiveRoleMaker', 'PaddleCloudRoleMaker', 'GeneralRoleMaker'
]


class Role:
    WORKER = 1
    SERVER = 2
    XPU = 3


class MockBarrier(object):
    """
    MockBarrier is a empty impletation for barrier
    mock as a real barrier for never-barrier in a specific scenario
    """

    def barrier(self):
        """
        dummy barrier, do nothing
        """
        pass

    def barrier_all(self):
        """
        dummy all barrier, do nothing
        """
        pass

    def all_reduce(self, obj):
        """
        dummy all reduce, do nothing
        Args:
            obj(any): obj to do all reduce
        """
        return obj

    def all_gather(self, obj):
        """
        dummy all gather, do nothing
        Args:
            obj(any): obj to do all gather
        """
        return [obj]


class RoleMakerBase(object):
    """
    RoleMakerBase is a base class for assigning a role to current process
    in distributed training.
    A paddle developer can implement RoleMakerBase to design a role maker
    for worker or pserver assignment.
    """

    def __init__(self):
        self._worker_endpoints = []
        self._server_endpoints = []
        self._role_is_generated = False
        self._role = None
        self._current_id = -1

    def is_worker(self):
        """
        return is_worker() of current process
        """
        raise NotImplementedError("Please implement this method in child class")

    def is_server(self):
        """
        return is_server() of current process
        """
        raise NotImplementedError("Please implement this method in child class")

    def is_first_worker(self):
        """
        Check whether the node is the first instance of worker.
        Returns:
            bool: True if this is the first node of worker,
                  False if not.
        """
        raise NotImplementedError("Please implement this method in child class")

    def worker_num(self):
        """
        Get current total worker number.

        Returns:
            int: worker number
        """
        raise NotImplementedError("Please implement this method in child class")

    def role_id(self):
        return self.worker_index() if self.is_worker() else self.server_index()

    def worker_index(self):
        """
        Get current worker id.

        Returns:
            int: node id
        """
        raise NotImplementedError("Please implement this method in child class")

    def server_index(self):
        """
        Get current server id.

        Returns:
            int: node id
        """
        raise NotImplementedError("Please implement this method in child class")

    def get_trainer_endpoints(self):
        """
        return trainer endpoints
        """
        return self._worker_endpoints

    def get_pserver_endpoints(self):
        """
        return pserver endpoints
        """
        return self._server_endpoints

    def to_string(self):
        return "role: {}, current_id: {}, worker_endpoints: {}, server_endpoints: {}".format(
            self._role, self._current_id, self._worker_endpoints,
            self._server_endpoints)

    def all_gather(self, input):
        """
        all gather between trainers and pservers

        Args:
            input(int|float): input value

        Returns:
            return a list of values
        """
        print("warning: RoleMakerBase does not have all gather.")
        return None

    def all_reduce_worker(self, input, output, mode="sum"):
        """
        all reduce between trainers if current role is TRAINER,
        only support array of one dim.

        Args:
            input(list/numpy.array): array of one dim
            output(list/numpy.array): array of one dim
            mode(str): "sum" or "min" or "max"
        """
        print("warning: RoleMakerBase does not have all reduce worker.")

    def barrier_worker(self):
        """
        barrier between trainers if current role is TRAINER
        """
        print("warning: RoleMakerBase does not have barrier worker.")

    def barrier_all(self):
        """
        barrier between trainers if current role is PSERVER
        """
        print("warning: RoleMakerBase does not have barrier all.")


class MPIRoleMaker(RoleMakerBase):
    """
    MPIRoleMaker is a MPI-API based role maker which is a counter-part of K8SRoleMaker
    mpi4py will be used if a developer inherits MPIRoleMaker
    """

    def __init__(self):
        """Init."""
        super(MPIRoleMaker, self).__init__()
        from mpi4py import MPI
        self.MPI = MPI
        self._comm = MPI.COMM_WORLD
        self._node_type_comm = None
        self._ips = None
        self._ip = None

    def _get_rank(self):
        """Return rank."""
        self._rank = self._comm.Get_rank()
        return self._rank

    def _get_size(self):
        """Return size."""
        self._size = self._comm.Get_size()
        return self._size

    def _all_gather(self, obj):
        """
        all_gather(obj) will call MPI's allgather function
        """
        self._barrier_all()
        return self._comm.allgather(obj)

    def _worker_gather(self, obj):
        """
        worker_gather(obj) will call MPI's allgather function
        """
        if self.is_worker():
            self._node_type_comm.barrier()
            return self._node_type_comm.allgather(obj)
        return None

    def _barrier_all(self):
        """
        barrier_all() will call MPI's barrier_all function
        """
        self._comm.barrier()

    def _finalize(self):
        """
        finalize the current MPI instance.
        """
        self.MPI.Finalize()

    def _get_ips(self):
        """
        collect current distributed job's ip list
        """
        if not self._ips:
            self._ips = self._comm.allgather(self.get_local_ip())
        return self._ips

    def get_local_ip(self):
        """Return get local ip."""
        import socket
        self._ip = socket.gethostbyname(socket.gethostname())
        return self._ip

    def generate_role(self):
        """
        generate_role() should be called to identify current process's role
        """
        raise NotImplementedError("Please implement this method in child class")


class MPISymetricRoleMaker(MPIRoleMaker):
    """
    MPISymetricRoleMaker is designed for worker and server assignment
    under MPI. Typically, a worker and a server node will be appointed
    on each physical node. This role maker can be only used under MPI.
    """

    def __init__(self):
        """Init."""
        super(MPISymetricRoleMaker, self).__init__()
        self._node_type = None
        self._proc_per_node = 2
        self._pserver_rand_port = 0

    def _check_role_generation(self):
        """Check whether role has been generated."""
        if not self._role_is_generated:
            raise NameError("generate_role() should be called first")
        return True

    def all_gather(self, input):
        """
        all gather between trainers and pservers

        Args:
            input(int|float): input value

        Returns:
            return a list of values
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._all_gather(input)

    def all_reduce_worker(self, input, output, mode="sum"):
        """
        all reduce between trainers if current role is TRAINER,
        only support array of one dim.

        Args:
            input(list/numpy.array): array of one dim
            output(list/numpy.array): array of one dim
            mode(str): "sum" or "min" or "max"
        """
        if not self._role_is_generated:
            self.generate_role()
        if not self.is_worker():
            print("warning: current role is not worker in all_reduce_worker")
            return
        self._all_reduce(input, output, mode)

    def barrier_worker(self):
        """
        barrier between trainers if current role is TRAINER
        """
        if not self._role_is_generated:
            self.generate_role()
        if self.is_worker():
            self._node_type_comm.barrier()
        else:
            print("warning: current role is not worker in barrier_worker")

    def barrier_all(self):
        """
        barrier between trainers if current role is PSERVER
        """
        if not self._role_is_generated:
            self.generate_role()
        self._comm.barrier()

    def is_first_worker(self):
        """
        return whether current process is the first worker assigned by role maker
        """
        if self._check_role_generation():
            return self.is_worker() and 0 == self.worker_index()
        return False

    def get_pserver_endpoints(self):
        """
        get pserver endpoints
        Returns:
            endpoints(list): pserver endpoints
        """
        if self._pserver_rand_port <= 0:
            import random
            random.seed(self._server_num())
            # port will be randomly generated from 60001 to 63999
            # random seed is server num so that all nodes will get
            # the same port
            self._pserver_rand_port = random.randint(60001, 64000)
        endpoints = [
            x + ":" + str(self._pserver_rand_port)
            for x in self._server_endpoints
        ]
        return endpoints

    def worker_num(self):
        return self._worker_num()

    def is_worker(self):
        """
        return whether current process is worker assigned by role maker
        """
        if self._check_role_generation():
            return self._node_type == 1
        return False

    def is_server(self):
        """
        return whether current process is server assigned by role maker
        """
        if self._check_role_generation():
            return self._node_type == 0
        return False

    def _worker_num(self):
        """
        return the current number of worker
        """
        if self._check_role_generation():
            return int(self._get_size() / self._proc_per_node)
        return 0

    def _server_num(self):
        """
        return the current number of server
        """
        if self._check_role_generation():
            return int(self._get_size() / self._proc_per_node)
        else:
            self.generate_role()
            return int(self._get_size() / self._proc_per_node)

    def worker_index(self):
        """
        return the index of worker
        """
        if self._check_role_generation():
            return int(self._rank / self._proc_per_node)
        else:
            self.generate_role()
            return int(self._get_size() / 2)

    def server_index(self):
        """
        return the index of server
        """
        if self._check_role_generation():
            return int(self._rank / self._proc_per_node)
        else:
            self.generate_role()
            return int(self._get_size() / self._proc_per_node)

    def _all_reduce(self, input, output, mode="sum"):
        """
        all reduce between trainers if current role is TRAINER,
        only support array of one dim.

        Args:
            input(list/numpy.array): array of one dim
            output(list/numpy.array): array of one dim
            mode(str): "sum" or "min" or "max"
        """
        if not self._role_is_generated:
            self.generate_role()
        if mode == "sum":
            mode = self.MPI.SUM
        elif mode == "max":
            mode = self.MPI.MAX
        elif mode == "min":
            mode = self.MPI.MIN
        else:
            raise ValueError("unknown mode: %s" % mode)
        self._node_type_comm.Allreduce(input, output, op=mode)

    def _barrier_worker(self):
        """
        barrier all workers in current distributed job
        """
        if self._check_role_generation():
            if self.is_worker():
                self._node_type_comm.barrier()
        else:
            raise Exception("You should check role generation first")

    def _barrier_server(self):
        """
        barrier all servers in current distributed job
        """
        if self._check_role_generation():
            if self.is_server():
                self._node_type_comm.barrier()
        else:
            raise Exception("You should check role generation first")

    def generate_role(self):
        """
        generate currently process's role
        """
        if not self._role_is_generated:
            # TODO(guru4elephant): only allow to be called once
            self._worker_endpoints = self._get_ips()[1::2]
            self._server_endpoints = self._get_ips()[::2]

            if 0 == self._get_rank() % self._proc_per_node % 2:
                self._node_type = 0
            else:
                self._node_type = 1
            self._node_type_comm = self._comm.Split(self._node_type)
            self._role_is_generated = True
        else:
            raise Exception("You should check role generation first")


class PaddleCloudRoleMaker(RoleMakerBase):
    """
    role maker for paddle cloud,
    base class is RoleMakerBase
    """

    def __init__(self, is_collective=False):
        super(PaddleCloudRoleMaker, self).__init__()
        self._role_is_generated = False
        self._is_collective = is_collective

    def generate_role(self):
        """Generate role."""
        if not self._role_is_generated:
            if not self._is_collective:
                try:
                    # Environment variable PADDLE_PSERVERS_IP_PORT_LIST must be set
                    # format: string(ip:port), eg. 127.0.0.1:6001
                    eplist = os.environ["PADDLE_PSERVERS_IP_PORT_LIST"].split(
                        ",")
                    # note that, we usually assign the same port to different ips
                    # if we run parameter server training in local mode
                    # port should be different in environment variables

                    trainers_num = int(os.environ["PADDLE_TRAINERS_NUM"])
                    training_role = os.environ["TRAINING_ROLE"]

                    if training_role not in ["TRAINER", "PSERVER"]:
                        raise ValueError(
                            "TRAINING_ROLE must be PSERVER or TRAINER")

                    if training_role == "TRAINER":
                        role = Role.WORKER
                        current_id = int(os.environ["PADDLE_TRAINER_ID"])
                    elif training_role == "PSERVER":
                        role = Role.SERVER
                        cur_ip = os.environ["POD_IP"]
                        curr_port = os.environ["PADDLE_PORT"]
                        curr_endpoint = ":".join([cur_ip, curr_port])
                        current_id = eplist.index(curr_endpoint)
                    else:
                        raise ValueError(
                            "TRAINING_ROLE must be PSERVER or TRAINER")
                except ValueError as ve:
                    raise ValueError(
                        "something wrong with PaddleCloud, please check environment"
                    )

                self._trainers_num = trainers_num
                self._server_endpoints = eplist
                self._role = role
                self._current_id = current_id
            else:
                self._current_id = int(os.getenv("PADDLE_TRAINER_ID", "0"))
                self._training_role = os.getenv("PADDLE_TRAINING_ROLE",
                                                "TRAINER")
                assert (self._training_role == "TRAINER")
                self._worker_endpoints = os.getenv("PADDLE_TRAINER_ENDPOINTS")
                self._current_endpoint = os.getenv("PADDLE_CURRENT_ENDPOINT")
                assert self._worker_endpoints is not None, "can't find PADDLE_TRAINER_ENDPOINTS"
                self._worker_endpoints = self._worker_endpoints.split(",")
                self._trainers_num = len(self._worker_endpoints)

            self._role_is_generated = True

    def get_pserver_endpoints(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._server_endpoints

    def is_worker(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.WORKER

    def is_server(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.SERVER

    def is_first_worker(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.WORKER and self._current_id == 0

    def worker_index(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._current_id

    def server_index(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._current_id

    def worker_num(self):
        if not self._role_is_generated:
            self.generate_role()
        return self._trainers_num


class GeneralRoleMaker(RoleMakerBase):
    """
    This role maker is for general use, you can set os.environ to customize:
        PADDLE_PSERVERS_IP_PORT_LIST : all pservers' ip:port, separated by ','
        PADDLE_TRAINER_ENDPOINTS     : all trainers' ip:port, separated by ','
        TRAINING_ROLE                : TRAINER or PSERVER
        PADDLE_TRAINER_ID            : current trainer id (only for trainer),
                                       it is index in PADDLE_TRAINER_ENDPOINTS
        PADDLE_PSERVER_ID            : current pserver id (only for pserver)
                                       it is index in PADDLE_PSERVERS_IP_PORT_LIST
    """

    def __init__(self, **kwargs):
        super(GeneralRoleMaker, self).__init__()
        self._role_is_generated = False
        self._hdfs_name = kwargs.get("hdfs_name", "")
        self._hdfs_ugi = kwargs.get("hdfs_ugi", "")
        self._hdfs_path = kwargs.get("path", "").rstrip("/")
        self._init_timeout_seconds = kwargs.get("init_timeout_seconds", 3600)
        self._run_timeout_seconds = kwargs.get("run_timeout_seconds", 9999999)
        self._use_metric = kwargs.get("use_metric", False)
        ip_port = kwargs.get("http_ip_port", "")
        self._use_ps_gpu = kwargs.get("use_ps_gpu", False)
        self._http_ip_port = []
        self._http_server = None
        # if ip_port is not empty, it will use http instead of hdfs
        if ip_port != "":
            self._http_ip_port = ip_port.split(":")
            # it's for communication between processes
            self._manager = Manager()
            # global dict to store status
            self._http_server_d = self._manager.dict()
            # set running status of http server
            self._http_server_d["running"] = False
        self._iface = self.__get_default_iface()
        self._iface = "" if self._iface == "lo" else self._iface
        # this environment variable can be empty
        self._prefix = os.getenv("SYS_JOB_ID", "")

    def generate_role(self):
        """
        generate role for general role maker
        """
        if not self._role_is_generated:
            eplist = os.environ["PADDLE_PSERVERS_IP_PORT_LIST"].split(",")
            training_role = os.environ["TRAINING_ROLE"]
            worker_endpoints = os.environ["PADDLE_TRAINER_ENDPOINTS"].split(",")
            trainers_num = len(worker_endpoints)
            if training_role not in ["TRAINER", "PSERVER"]:
                raise ValueError("TRAINING_ROLE must be PSERVER or TRAINER")
            self._is_barrier_all = 1
            if "PADDLE_IS_BARRIER_ALL_ROLE" in os.environ:
                self._is_barrier_all = int(
                    os.environ["PADDLE_IS_BARRIER_ALL_ROLE"])
            if training_role == "TRAINER":
                role = Role.WORKER
                current_id = int(os.environ["PADDLE_TRAINER_ID"])
                if current_id == 0 and len(self._http_ip_port) != 0:
                    size_d = {
                        "trainer": len(worker_endpoints),
                        "pserver": len(eplist),
                        "all": len(worker_endpoints) + len(eplist)
                    }
                    # child process for http server
                    self._http_server = Process(target=self.__start_kv_server,
                                                args=(self._http_server_d,
                                                      size_d))
                    self._http_server.daemon = True
                    # set running status to True
                    self._http_server_d["running"] = True
                    # start child process
                    self._http_server.start()
                self._node_type = 1
                self._cur_endpoint = worker_endpoints[current_id]
                if self._is_barrier_all:
                    gloo = fluid.core.Gloo()
                    gloo.set_rank(current_id)
                    gloo.set_size(len(worker_endpoints))
                    gloo.set_prefix(self._prefix)
                    gloo.set_iface(self._iface)
                    gloo.set_timeout_seconds(self._init_timeout_seconds,
                                             self._run_timeout_seconds)
                    if len(self._http_ip_port) != 0:
                        gloo.set_http_store(self._http_ip_port[0],
                                            int(self._http_ip_port[1]),
                                            "trainer")
                    else:
                        gloo.set_hdfs_store(self._hdfs_path + "/trainer",
                                            self._hdfs_name, self._hdfs_ugi)
                    gloo.init()
                    self._node_type_comm = gloo
                    if self._use_ps_gpu or self._use_metric:
                        Gloo_strategy = fluid.core.GlooParallelStrategy()
                        Gloo_strategy.rank = current_id
                        Gloo_strategy.rank_num = len(worker_endpoints)
                        Gloo_strategy.ip_address = self._http_ip_port[0]
                        Gloo_strategy.ip_port = int(self._http_ip_port[1])
                        Default_init_timeout_seconds = 3600
                        Default_run_timeout_seconds = 9999999
                        Gloo_strategy.init_seconds = Default_init_timeout_seconds
                        Gloo_strategy.run_seconds = Default_run_timeout_seconds
                        Gloo = fluid.core.GlooParallelContext(Gloo_strategy)
                        Gloo.init()
                else:
                    self._all_comm = MockBarrier()
            elif training_role == "PSERVER":
                role = Role.SERVER
                if os.environ.get("PADDLE_PSERVER_ID") is not None:
                    current_id = int(os.environ["PADDLE_PSERVER_ID"])
                    cur_endpoint = eplist[current_id]
                else:
                    # this is for compatible with paddlecloud
                    cur_ip = os.environ["POD_IP"]
                    cur_port = os.environ["PADDLE_PORT"]
                    cur_endpoint = ":".join([cur_ip, cur_port])
                    current_id = eplist.index(cur_endpoint)
                self._node_type = 0
                self._cur_endpoint = cur_endpoint
                gloo = fluid.core.Gloo()
                gloo.set_rank(current_id)
                gloo.set_size(len(eplist))
                gloo.set_prefix(self._prefix)
                gloo.set_iface(self._iface)
                gloo.set_timeout_seconds(self._init_timeout_seconds,
                                         self._run_timeout_seconds)
                if len(self._http_ip_port) != 0:
                    gloo.set_http_store(self._http_ip_port[0],
                                        int(self._http_ip_port[1]), "pserver")
                else:
                    gloo.set_hdfs_store(self._hdfs_path + "/pserver",
                                        self._hdfs_name, self._hdfs_ugi)
                gloo.init()
                self._node_type_comm = gloo

            gloo = fluid.core.Gloo()
            all_list = worker_endpoints + eplist
            gloo.set_rank(all_list.index(self._cur_endpoint))
            gloo.set_size(len(all_list))
            gloo.set_prefix(self._prefix)
            gloo.set_iface(self._iface)
            gloo.set_timeout_seconds(self._init_timeout_seconds,
                                     self._run_timeout_seconds)
            if len(self._http_ip_port) != 0:
                gloo.set_http_store(self._http_ip_port[0],
                                    int(self._http_ip_port[1]), "all")
            else:
                gloo.set_hdfs_store(self._hdfs_path + "/all", self._hdfs_name,
                                    self._hdfs_ugi)
            gloo.init()
            self._all_comm = gloo
            self._trainers_num = trainers_num
            self._server_endpoints = eplist
            self._role = role
            self._current_id = current_id
            self._rank = all_list.index(self._cur_endpoint)
            self._size = len(all_list)
            self._worker_endpoints = worker_endpoints
            self._role_is_generated = True

    def all_gather(self, input):
        """
        all gather between trainers and pservers

        Args:
            input(int|float): input value

        Returns:
            return a list of values
        """
        return self._all_gather(input)

    def all_reduce_worker(self, input, output, mode="sum"):
        """
        all reduce between trainers if current role is TRAINER,
        only support array of one dim.

        Args:
            input(list/numpy.array): array of one dim
            output(list/numpy.array): array of one dim
            mode(str): "sum" or "min" or "max"
        """
        if not self.is_worker():
            return
        self._all_reduce(input, output, mode)

    def barrier_worker(self):
        """
        barrier between trainers if current role is TRAINER
        """
        self._barrier_worker()

    def barrier_all(self):
        """
        barrier between trainers if current role is PSERVER
        """
        self._barrier_all()

    def get_local_endpoint(self):
        """
        get local endpoint of current process
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._cur_endpoint

    def get_trainer_endpoints(self):
        """
        get endpoint of all trainers
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._worker_endpoints

    def get_pserver_endpoints(self):
        """
        get endpoint of all pservers
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._server_endpoints

    def is_worker(self):
        """
        whether current process is worker
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.WORKER

    def is_server(self):
        """
        whether current process is server
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.SERVER

    def is_first_worker(self):
        """
        whether current process is worker of rank 0
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.WORKER and self._current_id == 0

    def worker_index(self):
        """
        get index of current worker
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._current_id

    def server_index(self):
        """
        get index of current server
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._current_id

    def worker_num(self):
        """
        retrun the current number of worker
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._worker_num()

    def server_num(self):
        """
        return the current number of server
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._server_num()

    def _barrier_worker(self):
        """
        barrier all workers in current distributed job
        """
        if not self._role_is_generated:
            self.generate_role()
        if self.is_worker():
            self._node_type_comm.barrier()

    def _barrier_all(self):
        """
        barrier all workers and servers in current distributed job
        """
        if not self._role_is_generated:
            self.generate_role()
        self._all_comm.barrier()

    def _barrier_server(self):
        """
        barrier all servers in current distributed job
        """
        if not self._role_is_generated:
            self.generate_role()
        if self.is_server():
            self._node_type_comm.barrier()

    def _worker_num(self):
        """
        return the current number of worker
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._trainers_num

    def _server_num(self):
        """
        return the current number of server
        """
        if not self._role_is_generated:
            self.generate_role()
        return len(self._server_endpoints)

    def _finalize(self):
        """Default do nothing."""
        pass

    def _all_reduce(self, input, output, mode="sum"):
        """
        all reduce between all workers

        Args:
            input(list|numpy.array): array of one dim
            output(list|numpy.array): array of one dim
            mode(str): "sum" or "min" or "max"
        """
        if not self._role_is_generated:
            self.generate_role()
        input_list = [i for i in input]
        ans = self._node_type_comm.all_reduce(input_list, mode)
        for i in range(len(ans)):
            output[i] = ans[i]

    def _all_gather(self, obj):
        """
        gather between all workers and pservers
        """
        if not self._role_is_generated:
            self.generate_role()
        self._barrier_all()
        return self._all_comm.all_gather(obj)

    def _worker_gather(self, obj):
        """
        gather between all workers
        """
        if not self._role_is_generated:
            self.generate_role()
        if not self.is_worker():
            return None
        self._barrier_worker()
        return self._node_type_comm.all_gather(obj)

    def _get_rank(self):
        """
        get current rank in all workers and pservers
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._rank

    def _get_size(self):
        """
        get total num of all workers and pservers
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._size

    def __get_default_iface(self):
        """
        get default physical interface
        """
        default1 = self.__get_default_iface_from_gateway()
        default2 = self.__get_default_iface_from_interfaces()
        return default2 if default1 == "lo" else default1

    def __get_default_iface_from_gateway(self):
        """
        get default physical interface
        """
        res = os.popen("route -A inet").read().strip().split("\n")

        gateway_idx = None
        iface_idx = None
        for item in res:
            item = item.split()
            if "Gateway" in item and "Iface" in item:
                gateway_idx = item.index("Gateway")
                iface_idx = item.index("Iface")
            elif gateway_idx != None and iface_idx != None:
                gateway = None
                if len(item) > gateway_idx:
                    gateway = item[gateway_idx]
                if gateway and gateway != '*' and gateway != "0.0.0.0" and len(
                        item) > iface_idx:
                    return item[iface_idx]
        return "lo"

    def __get_default_iface_from_interfaces(self):
        """
        get default physical interface
        """
        res = os.popen("ip -f inet addr | awk NR%3==1").read().strip().split(
            "\n")
        for item in res:
            if "BROADCAST" in item:
                return item.split(":")[1].strip()
        return "lo"

    def __start_kv_server(self, http_server_d, size_d):
        from paddle.fluid.incubate.fleet.utils.http_server import KVServer
        http_server = KVServer(int(self._http_ip_port[1]), size_d)
        http_server.start()
        wait_seconds = 5
        while http_server_d.get("running", False):
            time.sleep(wait_seconds)
        http_server.stop()


class HeterRoleMaker(GeneralRoleMaker):
    """
    This role maker is for general use, you can set os.environ to customize:
        PADDLE_PSERVERS_IP_PORT_LIST : all pservers' ip:port, separated by ','
        PADDLE_TRAINER_ENDPOINTS     : all trainers' ip:port, separated by ','
        TRAINING_ROLE                : TRAINER or PSERVER
        PADDLE_TRAINER_ID            : current trainer id (only for trainer),
                                       it is index in PADDLE_TRAINER_ENDPOINTS
        PADDLE_PSERVER_ID            : current pserver id (only for pserver)
                                       it is index in PADDLE_PSERVERS_IP_PORT_LIST
    """

    def generate_role(self):
        """
        generate role for general role maker
        """
        if not self._role_is_generated:
            eplist = os.environ["PADDLE_PSERVERS_IP_PORT_LIST"].split(",")
            training_role = os.environ["TRAINING_ROLE"]
            worker_endpoints = os.environ["PADDLE_TRAINER_ENDPOINTS"].split(",")
            trainers_num = len(worker_endpoints)
            xpu_endpoints = os.environ["PADDLE_XPU_ENDPOINTS"].split(",")
            xpu_num = len(xpu_endpoints)
            if training_role not in ["TRAINER", "PSERVER", "XPU"]:
                raise ValueError(
                    "TRAINING_ROLE must be PSERVER or TRAINER or XPU")
            if training_role == "TRAINER":
                role = Role.WORKER
                current_id = int(os.environ["PADDLE_TRAINER_ID"])
                self._node_type = 1
                self._cur_endpoint = worker_endpoints[current_id]
                gloo = fluid.core.Gloo()

                gloo.set_rank(current_id)
                gloo.set_size(len(worker_endpoints))
                gloo.set_prefix(self._prefix)
                gloo.set_iface(self._iface)
                gloo.set_timeout_seconds(self._init_timeout_seconds,
                                         self._run_timeout_seconds)
                gloo.set_hdfs_store(
                    self._hdfs_path.rstrip("/") + "/trainer", self._hdfs_name,
                    self._hdfs_ugi)
                gloo.init()
                self._node_type_comm = gloo
            elif training_role == "XPU":
                role = Role.XPU
                current_id = int(os.environ["PADDLE_XPU_ID"])
                self._node_type = 2
                self._cur_endpoint = xpu_endpoints[current_id]
                gloo = fluid.core.Gloo()

                gloo.set_rank(current_id)
                gloo.set_size(len(xpu_endpoints))
                gloo.set_prefix(self._prefix)
                gloo.set_iface(self._iface)
                gloo.set_timeout_seconds(self._init_timeout_seconds,
                                         self._run_timeout_seconds)
                gloo.set_hdfs_store(
                    self._hdfs_path.rstrip("/") + "/xpu", self._hdfs_name,
                    self._hdfs_ugi)
                gloo.init()
                self._node_type_comm = gloo
            elif training_role == "PSERVER":
                role = Role.SERVER
                if os.environ.get("PADDLE_PSERVER_ID") is not None:
                    current_id = int(os.environ["PADDLE_PSERVER_ID"])
                    cur_endpoint = eplist[current_id]
                else:
                    # this is for compatible with paddlecloud
                    cur_ip = os.environ["POD_IP"]
                    cur_port = os.environ["PADDLE_PORT"]
                    cur_endpoint = ":".join([cur_ip, cur_port])
                    current_id = eplist.index(cur_endpoint)
                self._node_type = 0
                self._cur_endpoint = cur_endpoint
                gloo = fluid.core.Gloo()
                gloo.set_rank(current_id)
                gloo.set_size(len(eplist))
                gloo.set_prefix(self._prefix)
                gloo.set_iface(self._iface)
                gloo.set_timeout_seconds(self._init_timeout_seconds,
                                         self._run_timeout_seconds)
                gloo.set_hdfs_store(
                    self._hdfs_path.rstrip("/") + "/pserver", self._hdfs_name,
                    self._hdfs_ugi)
                gloo.init()
                self._node_type_comm = gloo

            if training_role == "TRAINER" or training_role == "XPU":
                gloo = fluid.core.Gloo()
                heter_list = worker_endpoints + xpu_endpoints

                gloo.set_rank(heter_list.index(self._cur_endpoint))
                gloo.set_size(len(heter_list))
                gloo.set_prefix(self._prefix)
                gloo.set_iface(self._iface)
                gloo.set_timeout_seconds(self._init_timeout_seconds,
                                         self._run_timeout_seconds)
                gloo.set_hdfs_store(
                    self._hdfs_path.rstrip("/") + "/heter", self._hdfs_name,
                    self._hdfs_ugi)
                gloo.init()
                self._heter_comm = gloo

            gloo = fluid.core.Gloo()
            all_list = worker_endpoints + eplist + xpu_endpoints

            gloo.set_rank(all_list.index(self._cur_endpoint))
            gloo.set_size(len(all_list))
            gloo.set_prefix(self._prefix)
            gloo.set_iface(self._iface)
            gloo.set_timeout_seconds(self._init_timeout_seconds,
                                     self._run_timeout_seconds)
            gloo.set_hdfs_store(
                self._hdfs_path.rstrip("/") + "/all", self._hdfs_name,
                self._hdfs_ugi)
            gloo.init()

            self._all_comm = gloo
            self._trainers_num = trainers_num
            self._server_endpoints = eplist
            self._role = role
            self._current_id = current_id
            self._rank = all_list.index(self._cur_endpoint)
            self._size = len(all_list)
            self._worker_endpoints = worker_endpoints
            self._xpu_endpoints = xpu_endpoints
            self._role_is_generated = True

    def is_xpu(self):
        """
        whether current process is server
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.XPU

    def is_first_xpu(self):
        """
        whether current process is worker of rank 0
        """
        if not self._role_is_generated:
            self.generate_role()
        return self._role == Role.XPU and self._current_id == 0

    def _barrier_xpu(self):
        """
        barrier all workers in current distributed job
        """
        if not self._role_is_generated:
            self.generate_role()
        if self.is_xpu():
            self._node_type_comm.barrier()

    def _barrier_heter(self):
        """
        barrier all workers in current distributed job
        """
        if not self._role_is_generated:
            self.generate_role()
        if self.is_xpu() or self.is_worker:
            self._heter_comm.barrier()

    def xpu_num(self):
        """
        """
        if not self._role_is_generated:
            self.generate_role()
        return len(self._xpu_endpoints)


class UserDefinedRoleMaker(RoleMakerBase):
    """
    UserDefinedRoleMaker is designed for worker and server assignment
    under manual. Typically, a worker and a server node will be appointed
    on each physical node, It can be assign by user.
    """

    def __init__(self,
                 current_id=0,
                 role=Role.WORKER,
                 worker_num=0,
                 server_endpoints=None):
        super(UserDefinedRoleMaker, self).__init__()

        if not isinstance(server_endpoints, list):
            raise TypeError("server_endpoints must be as string list")
        elif len(server_endpoints) <= 0:
            raise ValueError(
                "the length of server_endpoints list must be greater than 0")
        elif len(server_endpoints) != len(set(server_endpoints)):
            raise ValueError("server_endpoints can't have duplicate elements")
        else:
            for server_endpoint in server_endpoints:
                if not isinstance(server_endpoint, str):
                    raise TypeError(
                        "every element in server_endpoints list must be as string"
                    )
            self._server_endpoints = server_endpoints

        if role != Role.WORKER and role != Role.SERVER:
            raise TypeError("role must be as Role")
        else:
            self._role = role

        if not isinstance(current_id, int):
            raise TypeError("current_id must be as int")
        else:
            if current_id < 0:
                raise ValueError(
                    "current_id must be greater than or equal to 0")
            elif self._role == Role.SERVER and current_id >= len(
                    server_endpoints):
                raise ValueError(
                    "if role is Role.SERVER, current_id must be less than or equal to len(server_endpoints) - 1"
                )
            self._current_id = current_id

        if not isinstance(worker_num, int):
            raise TypeError("worker_num must be as int")
        else:
            if worker_num <= 0:
                raise ValueError("worker_num must be greater than 0")
            self._worker_num = worker_num

    def generate_role(self):
        self._role_is_generated = True

    def is_worker(self):
        return self._role == Role.WORKER

    def is_server(self):
        return self._role == Role.SERVER

    def is_first_worker(self):
        return self._role == Role.WORKER and self._current_id == 0

    def worker_index(self):
        return self._current_id

    def server_index(self):
        return self._current_id

    def worker_num(self):
        return self._worker_num


class UserDefinedCollectiveRoleMaker(RoleMakerBase):
    """
    UserDefinedCollectiveRoleMaker is designed for worker assignment
    under manual for collective mode.
    """

    def __init__(self, current_id=0, worker_endpoints=None):
        super(UserDefinedCollectiveRoleMaker, self).__init__()

        if not isinstance(worker_endpoints, list):
            raise TypeError("worker_endpoints must be as string list")
        elif len(worker_endpoints) <= 0:
            raise ValueError(
                "the length of worker_endpoints list must be greater than 0")
        elif len(worker_endpoints) != len(set(worker_endpoints)):
            raise ValueError("worker_endpoints can't have duplicate elements")
        else:
            for worker_endpoint in worker_endpoints:
                if not isinstance(worker_endpoint, str):
                    raise TypeError(
                        "every element in worker_endpoints list must be as string"
                    )
            self._worker_endpoints = worker_endpoints

        if not isinstance(current_id, int):
            raise TypeError("current_id must be as int")
        else:
            if current_id < 0:
                raise ValueError(
                    "current_id must be greater than or equal to 0")
            elif current_id >= len(worker_endpoints):
                raise ValueError(
                    "current_id must be less than or equal to len(worker_endpoints) - 1"
                )
            self._current_id = current_id

        self._worker_num = len(self._worker_endpoints)

    def generate_role(self):
        self._role_is_generated = True

    def is_worker(self):
        return True

    def is_first_worker(self):
        return self._current_id == 0

    def worker_index(self):
        return self._current_id

    def worker_num(self):
        return self._worker_num