Bert镜像制作及flask生产环境模式启动

一天搞定两大技术点，成就满满。

一，dockerfile

FROM harbor.xxx.com.cn/3rd_part/tensorflow:1.14.0-gpu-py3-jupyter

LABEL "maintainer"="xxx4k"
LABEL "version"="1.0"

#COPY numpy-1.17.4-cp36-none-linux_x86_64.whl /tmp/
#COPY pyzmq-18.1.0-cp36-none-linux_x86_64.whl /tmp/

#RUN pip install /tmp/numpy-1.17.4-cp36-none-linux_x86_64.whl \
#    && pip install /tmp/pyzmq-18.1.0-cp36-none-linux_x86_64.whl \
RUN     pip install --no-cache-dir \
      -i http://xxx.com.cn/root/pypi/+simple/  \
      --trusted-host xxx.com.cn \
      tensorflow==1.14.0 bert-base==0.0.9 flask flask_compress flask_cors  flask_json \
    && rm -rf /tmp/* \
    && rm -rf ~/.cache/pip \
    && echo "finished"

二，修改Http.py

参考URL：
https://www.jianshu.com/p/beab4df088df
https://blog.csdn.net/jusang486/article/details/82382358
https://blog.csdn.net/AbeBetter/article/details/77652457
https://blog.csdn.net/anh3000/article/details/83047027

如果在flask里，使用app.run()的模式，输出总会提示：

* Serving Flask app "bert_base.server.http" (lazy loading)
* Environment: production
WARNING: This is a development server. Do not use it in a production deployment.
Use a production WSGI server instead.
* Debug mode: off
I1113 02:44:05.755911 139926823548672 _internal.py:122] * Running on http://0.0.0.0:8091/ (Press CTRL+C to quit)

那如何改进呢？
可以选用nginx或是tornado。
如果是代码模式下，tornador是首选。

from multiprocessing import Process
from tornado.wsgi import WSGIContainer
from tornado.httpserver import HTTPServer
from tornado.ioloop import IOLoop
import asyncio
from termcolor import colored

from .helper import set_logger


class BertHTTPProxy(Process):
    def __init__(self, args):
        super().__init__()
        self.args = args

    def create_flask_app(self):
        try:
            from flask import Flask, request
            from flask_compress import Compress
            from flask_cors import CORS
            from flask_json import FlaskJSON, as_json, JsonError
            from bert_base.client import ConcurrentBertClient
        except ImportError:
            raise ImportError('BertClient or Flask or its dependencies are not fully installed, '
                              'they are required for serving HTTP requests.'
                              'Please use "pip install -U bert-serving-server[http]" to install it.')

        # support up to 10 concurrent HTTP requests
        bc = ConcurrentBertClient(max_concurrency=self.args.http_max_connect,
                                  port=self.args.port, port_out=self.args.port_out,
                                  output_fmt='list', mode=self.args.mode)
        app = Flask(__name__)
        logger = set_logger(colored('PROXY', 'red'))

        @app.route('/status/server', methods=['GET'])
        @as_json
        def get_server_status():
            return bc.server_status

        @app.route('/status/client', methods=['GET'])
        @as_json
        def get_client_status():
            return bc.status

        @app.route('/encode', methods=['POST'])
        @as_json
        def encode_query():
            data = request.form if request.form else request.json
            try:
                logger.info('new request from %s' % request.remote_addr)
                print(data)
                return {'id': data['id'],
                        'result': bc.encode(data['texts'], is_tokenized=bool(
                            data['is_tokenized']) if 'is_tokenized' in data else False)}

            except Exception as e:
                logger.error('error when handling HTTP request', exc_info=True)
                raise JsonError(description=str(e), type=str(type(e).__name__))

        CORS(app, origins=self.args.cors)
        FlaskJSON(app)
        Compress().init_app(app)
        return app

    def run(self):
        app = self.create_flask_app()
        # app.run(port=self.args.http_port, threaded=True, host='0.0.0.0')
        # tornado 5 中引入asyncio.set_event_loop即可
        asyncio.set_event_loop(asyncio.new_event_loop())
        http_server = HTTPServer(WSGIContainer(app))
        http_server.listen(self.args.http_port)
        IOLoop.instance().start()

三，启动命令

bert-base-serving-start -bert_model_dir "/bert/outputFile_pb" -model_dir "/bert/outputFile_pb" -model_pb_dir "/bert/outputFile_pb" -mode CLASS -max_seq_len 64 -http_port 8091 -port 5575 -port_out 5576 

扫描二维码关注公众号，回复： 7845962 查看本文章