test_dist_fleet_base.py

#   Copyright (c) 2018 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.

"""
high level unit test for distribute fleet.
"""

import argparse
import os
import shutil
import socket
import subprocess
import sys
import tempfile
import time
import unittest
from contextlib import closing

import paddle
import paddle.distributed.fleet as fleet
import paddle.distributed.fleet.base.role_maker as role_maker
import paddle.fluid as fluid
from paddle.distributed.fleet.utils.ps_util import DistributedInfer

paddle.enable_static()

__all__ = ['FleetDistRunnerBase', 'TestFleetBase', 'runtime_main']

RUN_STEP = 5
LEARNING_RATE = 0.01
DIST_UT_PORT = 0


class FleetDistRunnerBase:
    """
    run_pserver,run_trainer : after init role, using transpiler split program
    net : implment by child class, the network of model
    do training : exe run program
    """

    def __init__(self):
        self._exe = None

    def build_role(self, args):

        if args.role.upper() == "PSERVER":
            role = role_maker.UserDefinedRoleMaker(
                is_collective=False,
                init_gloo=False,
                path=args.gloo_path,
                current_id=args.current_id,
                role=role_maker.Role.SERVER,
                worker_endpoints=args.trainer_endpoints.split(","),
                server_endpoints=args.endpoints.split(","),
            )
        else:
            role = role_maker.UserDefinedRoleMaker(
                is_collective=False,
                init_gloo=False,
                path=args.gloo_path,
                current_id=args.current_id,
                role=role_maker.Role.WORKER,
                worker_endpoints=args.trainer_endpoints.split(","),
                server_endpoints=args.endpoints.split(","),
            )
        self.role = role
        return role

    def build_strategy(self, args):
        if args.mode == "sync":
            self.strategy = paddle.distributed.fleet.DistributedStrategy()
            self.strategy.a_sync = False
        elif args.mode == "async":
            self.strategy = paddle.distributed.fleet.DistributedStrategy()
            self.strategy.a_sync = True
        elif args.mode == "geo":
            self.strategy = paddle.distributed.fleet.DistributedStrategy()
            self.strategy.a_sync = True
            self.strategy.a_sync_configs = {
                "k_steps": args.geo_sgd_need_push_nums
            }
        elif args.mode == "auto":
            self.strategy = paddle.distributed.fleet.DistributedStrategy()
            self.strategy.auto = True

        self.dump_param = os.getenv("dump_param", "").split(",")
        self.dump_fields = os.getenv("dump_fields", "").split(",")
        self.dump_fields_path = os.getenv("dump_fields_path", "")
        debug = int(os.getenv("Debug", "0"))
        # TODO(update strategy to support dump params)
        if False:  # debug:
            self.strategy.set_debug_opt(
                {
                    "dump_param": self.dump_param,
                    "dump_fields": self.dump_fields,
                    "dump_fields_path": self.dump_fields_path,
                }
            )

        return self.strategy

    def build_optimizer(self, avg_cost, strategy):
        use_grad_clip = int(os.getenv('GRAD_CLIP', 0))
        grad_clip = None
        if use_grad_clip:
            # 1: clip_by_value; 2: clip_by_norm; 3:clip_by_global_norm
            if use_grad_clip == 1:
                grad_clip = paddle.nn.ClipGradByValue(min=-5.0, max=5.0)
            elif use_grad_clip == 2:
                grad_clip = paddle.nn.ClipGradByNorm(2.0)
            elif use_grad_clip == 3:
                grad_clip = paddle.nn.ClipGradByGlobalNorm(2.0)

        use_decay = int(os.getenv("USE_DECAY", "0"))
        if use_decay:
            scheduler = paddle.optimizer.lr.ExponentialDecay(
                learning_rate=LEARNING_RATE, gamma=0.999, verbose=True
            )
            optimizer = fluid.optimizer.SGD(scheduler, grad_clip=grad_clip)
            """
            # learning rate decay method before 2.0
            optimizer = fluid.optimizer.SGD(
                learning_rate=fluid.layers.exponential_decay(
                    learning_rate=LEARNING_RATE,
                    decay_steps=500,
                    decay_rate=0.969,
                    staircase=True))
            """
        else:
            optimizer = fluid.optimizer.SGD(LEARNING_RATE, grad_clip=grad_clip)
        optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
        optimizer.minimize(avg_cost)

    def run_pserver(self, args):
        fleet.init_server()
        fleet.run_server()

    def run_dataset_trainer(self, args):
        out = self.do_dataset_training(fleet)

    def run_pyreader_trainer(self, args):
        out = self.do_pyreader_training(fleet)

    def net(self, args, batch_size=4, lr=0.01):
        raise NotImplementedError(
            "get_model should be implemented by child classes."
        )

    def get_executor(self):
        if self._exe is None:
            device_env = os.getenv("DEVICE", 'cpu')
            if device_env == 'cpu':
                device = fluid.CPUPlace()
            elif device_env == 'gpu':
                device = fluid.CUDAPlace(0)
            self._exe = fluid.Executor(device)
        return self._exe

    def do_dataset_training(self, fleet):
        raise NotImplementedError(
            "do_dataset_training should be implemented by child classes."
        )

    def do_pyreader_training(self, fleet):
        raise NotImplementedError(
            "do_pyreader_training should be implemented by child classes."
        )

    def do_distributed_testing(self, fleet):
        raise NotImplementedError(
            "do_distributed_testing should be implemented by child classes."
        )


class TestFleetBase(unittest.TestCase):
    """
    start_pserver,start_trainer : add start cmd to test
    run_cluster : using multi process to test distribute program
    """

    def _setup_config(self):
        raise NotImplementedError("tests should have _setup_config implemented")

    def tearDown(self):
        t = time.time() - self.startTime
        print('%s: %.3f' % (self.__class__.__name__, t))

    def setUp(self):
        self.startTime = time.time()

        self._mode = "sync"
        self._reader = "pyreader"
        self._trainers = 2
        self._pservers = 2
        self._need_test = 0
        self._model_dir = ""
        self._port_set = set()

        global DIST_UT_PORT
        if DIST_UT_PORT == 0 and os.getenv("PADDLE_DIST_UT_PORT"):
            DIST_UT_PORT = int(os.getenv("PADDLE_DIST_UT_PORT"))

        if DIST_UT_PORT:
            print("set begin_port:", DIST_UT_PORT)
            self._ps_endpoints = "127.0.0.1:%s,127.0.0.1:%s" % (
                DIST_UT_PORT,
                DIST_UT_PORT + 1,
            )
            self._tr_endpoints = "127.0.0.1:%s,127.0.0.1:%s" % (
                DIST_UT_PORT + 2,
                DIST_UT_PORT + 3,
            )
            DIST_UT_PORT += 4
        else:
            self._ps_endpoints = "127.0.0.1:%s,127.0.0.1:%s" % (
                self._find_free_port(),
                self._find_free_port(),
            )
            self._tr_endpoints = "127.0.0.1:%s,127.0.0.1:%s" % (
                self._find_free_port(),
                self._find_free_port(),
            )

        self._python_interp = sys.executable
        self._geo_sgd_need_push_nums = 5
        self._grad_clip_mode = 0
        self._setup_config()

    def _find_free_port(self):
        def __free_port():
            with closing(
                socket.socket(socket.AF_INET, socket.SOCK_STREAM)
            ) as s:
                s.bind(('', 0))
                return s.getsockname()[1]

        while True:
            port = __free_port()
            if port not in self._port_set:
                self._port_set.add(port)
                return port

    def _start_pserver(self, cmd, required_envs):
        ps0_cmd, ps1_cmd = cmd.format(0), cmd.format(1)

        log_dirname = required_envs.get("LOG_DIRNAME", tempfile.gettempdir())
        log_prename = required_envs.get("LOG_PREFIX", "")

        if log_dirname:
            log_prename += "_"

        ps0_err_log = os.path.join(log_dirname, log_prename + "ps0_stderr.log")
        ps1_err_log = os.path.join(log_dirname, log_prename + "ps1_stderr.log")
        ps0_out_log = os.path.join(log_dirname, log_prename + "ps0_stdout.log")
        ps1_out_log = os.path.join(log_dirname, log_prename + "ps1_stdout.log")

        ps0_err = open(ps0_err_log, "wb+")
        ps1_err = open(ps1_err_log, "wb+")

        ps0_out = open(ps0_out_log, "wb+")
        ps1_out = open(ps1_out_log, "wb+")

        ps0_proc = subprocess.Popen(
            ps0_cmd.strip().split(" "),
            stdout=ps0_out,
            stderr=ps0_err,
            env=required_envs,
        )

        ps1_proc = subprocess.Popen(
            ps1_cmd.strip().split(" "),
            stdout=ps1_out,
            stderr=ps1_err,
            env=required_envs,
        )

        return (
            (ps0_proc, ps0_out, ps0_err, ps0_out_log, ps0_err_log),
            (ps1_proc, ps1_out, ps1_err, ps1_out_log, ps1_err_log),
        )

    def _start_trainer(self, cmd, required_envs):
        tr0_cmd, tr1_cmd = cmd.format(0), cmd.format(1)

        log_dirname = required_envs.get("LOG_DIRNAME", tempfile.gettempdir())
        log_prename = required_envs.get("LOG_PREFIX", "")

        if log_dirname:
            log_prename += "_"

        tr0_err_log = os.path.join(log_dirname, log_prename + "tr0_stderr.log")
        tr1_err_log = os.path.join(log_dirname, log_prename + "tr1_stderr.log")
        tr0_out_log = os.path.join(log_dirname, log_prename + "tr0_stdout.log")
        tr1_out_log = os.path.join(log_dirname, log_prename + "tr1_stdout.log")

        tr0_err = open(tr0_err_log, "wb+")
        tr1_err = open(tr1_err_log, "wb+")

        tr0_out = open(tr0_out_log, "wb+")
        tr1_out = open(tr1_out_log, "wb+")

        tr0_proc = subprocess.Popen(
            tr0_cmd.strip().split(" "),
            stdout=tr0_out,
            stderr=tr0_err,
            env=required_envs,
        )

        tr1_proc = subprocess.Popen(
            tr1_cmd.strip().split(" "),
            stdout=tr1_out,
            stderr=tr1_err,
            env=required_envs,
        )

        return (
            (tr0_proc, tr0_out, tr0_err, tr0_out_log, tr0_err_log),
            (tr1_proc, tr1_out, tr1_err, tr1_out_log, tr1_err_log),
        )

    def _run_cluster(self, model, envs):
        env = {'GRAD_CLIP': str(self._grad_clip_mode), 'WITH_DISTRIBUTE': 'ON'}
        python_path = self._python_interp
        gloo_path = tempfile.mkdtemp()

        if os.getenv('WITH_COVERAGE', 'OFF') == 'ON':
            envs['COVERAGE_FILE'] = os.getenv('COVERAGE_FILE', '')
            python_path += " -m coverage run --branch -p"
        env.update(envs)

        tr_cmd = "{0} {1} --role trainer --endpoints {2} --trainer_endpoints {3} --current_id {{}} --trainers {4} --mode {5} --geo_sgd_need_push_nums {6} --reader {7} --gloo_path {8} --test {9}".format(
            python_path,
            model,
            self._ps_endpoints,
            self._tr_endpoints,
            self._trainers,
            self._mode,
            self._geo_sgd_need_push_nums,
            self._reader,
            gloo_path,
            self._need_test,
        )

        ps_cmd = "{0} {1} --role pserver --endpoints {2} --trainer_endpoints {3} --current_id {{}} --trainers {4} --mode {5} --geo_sgd_need_push_nums {6} --reader {7} --gloo_path {8} --test {9}".format(
            python_path,
            model,
            self._ps_endpoints,
            self._tr_endpoints,
            self._trainers,
            self._mode,
            self._geo_sgd_need_push_nums,
            self._reader,
            gloo_path,
            self._need_test,
        )

        if self._model_dir:
            tr_cmd += " --model_dir {}".format(self._model_dir)
            ps_cmd += " --model_dir {}".format(self._model_dir)

        # Run dist train to compare with local results
        ps0, ps1 = self._start_pserver(ps_cmd, env)
        tr0, tr1 = self._start_trainer(tr_cmd, env)

        ps0_proc, ps0_out, ps0_err, ps0_out_log, ps0_err_log = ps0
        ps1_proc, ps1_out, ps1_err, ps1_out_log, ps1_err_log = ps1

        tr0_proc, tr0_out, tr0_err, tr0_out_log, tr0_err_log = tr0
        tr1_proc, tr1_out, tr1_err, tr1_out_log, tr1_err_log = tr1

        # Wait until trainer process terminate
        # time_out = 120
        time_out = 60
        cur_time = 0

        while True:
            stat0 = tr0_proc.poll()
            stat1 = tr1_proc.poll()

            if stat0 is not None and stat1 is not None:
                break
            else:
                time.sleep(0.5)
                cur_time += 0.5

            if cur_time >= time_out:
                tr0_proc.terminate()
                tr1_proc.terminate()
                tr0_proc.wait()
                tr1_proc.wait()
                break

        tr0_ret = tr0_proc.returncode
        tr1_ret = tr1_proc.returncode

        ps0_proc.kill()
        ps1_proc.kill()
        ps0_proc.wait()
        ps1_proc.wait()

        def is_listen_failed(logx):
            is_lf = False

            listen_rgx = "Fail to listen"

            with open(logx, "r") as rb:
                for line in rb.readlines():
                    if listen_rgx in line:
                        is_lf = True
                        break
            return is_lf

        def catlog(logx):
            basename = os.path.basename(logx)
            print(
                "\n================== Error {} begin =====================".format(
                    basename
                )
            )
            os.system("cat {}".format(logx))
            print(
                "================== Error {} end =====================\n".format(
                    basename
                )
            )

        if tr0_ret != 0 or tr1_ret != 0:
            if is_listen_failed(ps0_err) or is_listen_failed(ps1_err):
                print("find parameter server port bind failed, skip the error")
                tr0_ret, tr1_ret = 0, 0
            else:
                for out, err in [
                    (ps0_out_log, ps0_err_log),
                    (ps1_out_log, ps1_err_log),
                    (tr0_out_log, tr0_err_log),
                    (tr1_out_log, tr1_err_log),
                ]:
                    catlog(out)
                    catlog(err)

        for pipe in [
            tr0_err,
            tr0_out,
            tr1_err,
            tr1_out,
            ps0_err,
            ps0_out,
            ps1_err,
            ps1_out,
        ]:
            pipe.close()

        shutil.rmtree(gloo_path)

        self.assertEqual(tr0_ret, 0, "something wrong in tr0, please check")
        self.assertEqual(tr1_ret, 0, "something wrong in tr1, please check")

        return 0, 0

    def check_with_place(
        self, model_file, delta=1e-3, check_error_log=False, need_envs={}
    ):
        required_envs = {
            "PATH": os.getenv("PATH", ""),
            "PYTHONPATH": os.getenv("PYTHONPATH", ""),
            "LD_LIBRARY_PATH": os.getenv("LD_LIBRARY_PATH", ""),
            "FLAGS_rpc_deadline": "5000",  # 5sec to fail fast
            "http_proxy": "",
        }

        required_envs.update(need_envs)

        if check_error_log:
            required_envs["GLOG_v"] = "3"
            required_envs["GLOG_logtostderr"] = "1"

        tr0_losses, tr1_losses = self._run_cluster(model_file, required_envs)


def runtime_main(test_class):
    parser = argparse.ArgumentParser(description='Run Fleet test.')
    parser.add_argument(
        '--role', type=str, required=True, choices=['pserver', 'trainer']
    )
    parser.add_argument('--endpoints', type=str, required=False, default="")
    parser.add_argument(
        '--trainer_endpoints', type=str, required=False, default=""
    )
    parser.add_argument('--gloo_path', type=str, required=False, default="")
    parser.add_argument('--current_id', type=int, required=False, default=0)
    parser.add_argument('--trainers', type=int, required=False, default=1)
    parser.add_argument('--mode', type=str, required=False, default='geo')
    parser.add_argument(
        '--geo_sgd_need_push_nums', type=int, required=False, default=2
    )
    parser.add_argument('--reader', type=str, required=False, default='dataset')
    parser.add_argument('--test', type=int, required=False, default=0)
    parser.add_argument('--model_dir', type=str, required=False, default="")
    args = parser.parse_args()

    model = test_class()
    role = model.build_role(args)

    # for distributed inference
    if args.test and args.model_dir != "":
        avg_cost = model.net(args, is_train=False)
        dist_infer = DistributedInfer()
        dist_infer.init_distributed_infer_env(
            exe=model.get_executor(),
            loss=model.avg_cost,
            role_maker=role,
            dirname=args.model_dir,
        )

        if fleet.is_worker():
            with paddle.static.program_guard(
                main_program=dist_infer.get_dist_infer_program()
            ):
                model.do_distributed_testing(fleet)
                fleet.stop_worker()
            return

        if fleet.is_server():
            return

    fleet.init(role)
    strategy = model.build_strategy(args)
    avg_cost = model.net(args)
    model.build_optimizer(avg_cost, strategy)

    if args.role == "pserver":
        model.run_pserver(args)
    else:
        if args.reader == "dataset":
            model.run_dataset_trainer(args)
        else:
            model.run_pyreader_trainer(args)

        if args.test:
            test_origin_program = paddle.static.Program()
            test_startup_program = paddle.static.Program()
            with paddle.static.program_guard(
                main_program=test_origin_program,
                startup_program=test_startup_program,
            ):
                with paddle.utils.unique_name.guard():
                    avg_cost = model.net(args, is_train=False)
            dist_infer = DistributedInfer(
                main_program=test_origin_program,
                startup_program=test_startup_program,
            )
            with paddle.static.program_guard(
                main_program=dist_infer.get_dist_infer_program()
            ):
                model.do_distributed_testing(fleet)
        fleet.stop_worker()