我使用tornado构建我的Web服务，并使用logging来获取tornado的记录器，一切看起来都很成功。但由于服务是multiprocessing，所以今天，当我检查日志时，我发现一些信息丢失了。所以我想问一下，如果我为不同的进程打开不同的日志，这个问题可以解决吗？
或者，如果在输出日志时可以将任何其他解决方案应用于multi-processes服务器。
这是我在我的解决方案中使用的代码：
https://gist.github.com/hcl14/259432dd648180bf2af672c26d9df9fc
它同时运行带有烧瓶应用程序的龙卷风服务器，登录屏幕，进入公共日志（stdout.log）和特定于进程的日志。只需在代码中查看注释，如何导入和组织所有内容。
要运行所有内容，将所有三个文件放在一起，创建logs子文件夹，运行mainprogram.py并测试发送不同的POST请求。然后，您可以检查日志文件夹中的文件，以确保正确记录所有内容。
mainprogram.py是一个主文件，其中logger被初始化并放入global_vars模块。必须在任何地方导入此模块，并从那里派生特定于流程的记录器。原因是这只是跨模块在全局范围内存储变量的便捷方式。当新进程分叉时，它会使用特定于进程的记录器覆盖记录器，因此进程使用的所有模块都将写入相应的记录器：
separate_logging.py（为方便起见，将其改为mainprogram.py）：
import logging
import os
import multiprocessing
from tornado.wsgi import WSGIContainer
from tornado.httpserver import HTTPServer
from tornado.ioloop import IOLoop
from tornado.options import define, options
# a way to pass variables into separate modules,
# process-specific because of 'fork' mode
# (each process will have its own version of this module)
import global_vars
logPath = os.environ.get('LOGFOLDER','logs')
fileName = os.environ.get('LOGFILE', "stdout.log")
address = os.environ.get('ADDRESS','0.0.0.0')
port = os.environ.get('PORT','8888')
NUM_PROCESSES = os.cpu_count()
# initializes the main logger to be used across all modules
logFormatter = logging.Formatter("%(asctime)s [%(processName)-12.12s] [%(threadName)-12.12s] [%(levelname)-5.5s] [%(filename)s:%(lineno)d] %(message)s")
rootLogger = logging.getLogger(__name__)
# first handler is general log
fileHandler = logging.FileHandler("{0}/{1}".format(logPath, fileName))
fileHandler.setFormatter(logFormatter)
rootLogger.addHandler(fileHandler)
# second handler is logging to console
consoleHandler = logging.StreamHandler()
consoleHandler.setFormatter(logFormatter)
rootLogger.addHandler(consoleHandler)
rootLogger.setLevel("DEBUG") # log everything
rootLogger.propagate = False
# until process branches, it uses rootLogger
global_vars.multiprocess_globals["logger"] = rootLogger
# third handler is process-specific log, initialized in processes
def init_logger2(secondary_logfile, rootLogger):
fileHandler1 = logging.FileHandler("{0}/{1}".format(logPath, 'process_'+str(secondary_logfile)+'.log'))
fileHandler1.setFormatter(logFormatter)
rootLogger.addHandler(fileHandler1)
return rootLogger
# external modules import goes here!
# otherwise they will not find any logger!
from flask_app import create_app
# ---------------
# process function
def run(process_id):
# initialize process-specific logger
processLogger = init_logger2(process_id, rootLogger)
global_vars.multiprocess_globals["logger"] = processLogger
# here you can run tornado app:
try:
app = create_app() # pass interests to flask app
ioloop = IOLoop()
http_server_api = HTTPServer(WSGIContainer(app))
# reuse_port allows multiple servers co-exist
# as separate processes
http_server_api.bind(address=address, port=port, reuse_port=True)
http_server_api.start()
processLogger.info("Process %s started %s:%s" % (process_id, address,
port))
ioloop.start()
except Exception as e:
processLogger.error(e)
# start processes (tornado servers)
if __name__ == '__main__':
processes = []
for i in range(1,NUM_PROCESSES):
p = multiprocessing.Process(target=run, args=(str(i),))
p.daemon = False # if we want to spawn child processes
#p.daemon = True # if we want to gracefully stop program
processes.append(p)
# Run processes:
for p in processes:
p.start()
# block program from exiting
for p in processes:
p.join()
global_vars.py：
# global variables to be used across processes.
# Separate file is needed to make globals accessible from different submodules.
# Global variables which need to exist in the scope of each process.
# Variables are added into this dictionary during process initialization.
multiprocess_globals = {}
flask_app.py：只是一个烧瓶中的应用程序，它使用更深层次的模块。在这种情况下，这些模块也应该像flask_app那样导入logger：
# create flask app to be run by tornado process
# process-specific globals
import global_vars
from flask import Flask, request, Response, json, abort, jsonify
import json as json2
# from deeper_module import do_something
app = Flask(__name__)
app.config.from_object(__name__)
# get process-specific logger
# do such import in any submodule !
# as global_vars is changed on process fork!
rootLogger = global_vars.multiprocess_globals["logger"]
logger = rootLogger.getChild(__name__)
def create_app():
app = Flask(__name__)
@app.route('/my_url1', methods=['POST'])
def my_url1():
body = json.loads(request.data)
# debug to process-specific logger
logger.debug(body)
# do_something()
response = app.response_class(
response=json.dumps({'response':'good'}),
status=200,
mimetype='application/json'
)
return response
# another functions
return app
在主程序中初始化global_vars之后，必须导入所有使用特定记录器的模块。启动程序后，您将看到：
2018-10-16 12:50:20,988 [Process-1 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 1 started 0.0.0.0:8888
2018-10-16 12:50:20,989 [Process-2 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 2 started 0.0.0.0:8888
2018-10-16 12:50:20,990 [Process-3 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 3 started 0.0.0.0:8888
2018-10-16 12:50:20,991 [Process-4 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 4 started 0.0.0.0:8888
2018-10-16 12:50:20,991 [Process-6 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 6 started 0.0.0.0:8888
2018-10-16 12:50:20,992 [Process-7 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 7 started 0.0.0.0:8888
2018-10-16 12:50:20,993 [Process-5 ] [MainThread ] [INFO ] [separate_logging.py:86] Process 5 started 0.0.0.0:8888
然后，您可以触发POST请求，该请求将由其中一个已启动的进程处理：
$ curl -H "Content-Type: application/json" -X POST -d '{"bla-bla":"bla"}' 127.0.0.1:8888/my_url1
结果是：
2018-10-16 12:51:40,040 [Process-5 ] [MainThread ] [DEBUG] [flask_app.py:31] {'bla-bla': 'bla'}
您可以检查此行是否也会出现在常规日志logs/stdout.log和logs/process_5.log中。