子进程阻止Django视图 [英] subprocess blocks Django view

问题描述

从视图调用subprocess.Popen时遇到问题:子过程完成之前，不会显示调用subprocess.Popen的视图.服务器立即发送"200 OK"，但不发送页面内容.

我的问题是:这是Django开发服务器的限制还是我做错了?

服务器无法完全挂起，因为可以同时处理其他视图.

已经有有关该主题的一些问题，Google给出了

如何繁殖

创建测试项目&应用:

cd/tmp
django-admin.py startproject django_test
cd django_test
./manage.py startapp subprocess_test

替换urls.py&subprocess_test/views.py，其中:

• urls.py:

  从django.conf.urls.defaults导入*

  urlpatterns =模式(''，
   (r'^ hello $'，'subprocess_test.views.hello')，
   (r'^ start $'，'subprocess_test.views.start')，
  )

• subprocess_test/views.py

  从django.http导入HttpResponse

  导入子流程

  def hello(request):
   返回HttpResponse('Hello world！')

  def start(请求):
   subprocess.Popen(["/bin/sleep"，"10"])
   返回HttpResponse('开始完成')

对其进行测试:

./manage.py运行服务器0.0.0.0:8000

转到 http://127.0.0.1:8000/hello 和

[2011年2月1日07:20:57]"GET/hello HTTP/1.1" 200 12
[01/Feb/2011 07:21:01]"GET/start HTTP/1.1" 200 10
[01/Feb/2011 07:21:01]"GET/hello HTTP/1.1" 200 12
[01/Feb/2011 07:21:02]"GET/hello HTTP/1.1" 200 12

使用wget:

wget http://127.0.0.1:8000/start
--2011-02-01 14:31:11-- http://127.0.0.1:8000/start
正在连接至127.0.0.1:8000 ...已连接.
HTTP请求已发送，正在等待响应... 200 OK
长度:未指定[text/html]
保存到:开始"

  [< =>]在9,5秒内达到10 --.- K/s 

2011-02-01 14:31:21(1,05 B/s)-«开始»保存[10]

解决方案

似乎您并不在乎系统调用的结果，所以我假设您正在尝试进行某种脱机(或后台操作))处理.

我建议采用一种更清洁的方式进行处理，而不是直接执行程序.使用诸如 Gearman 之类的排队系统来排队处理任务，然后让一个单独的工作程序从该队列中消费项目./p>

这样做的好处是可以在出现大流量高峰时保护服务器，因此您不必在每次向该视图发出请求时都派生一个进程.您可以根据自己的决定消费速度快慢，而不受流量的影响.

交通问题可能不是问题，但我个人认为这也是一个更清洁的设计决定.

I have a problem calling subprocess.Popen from a view: The view that calls subprocess.Popen is not displayed until the subprocess finishes. The server send "200 OK" immediately, but not the content of the page.

My question is: Is this a limitation of Django's development server or am I doing it wrong?

The server does not completely hangs, as other views can be processed in the meantime.

There are already a few questions on that topic and Google gives a few other threads, but I cannot find a clear answer to my question.

I believe this is not a python issue as this commands terminate immediately:

python -c 'import subprocess; print subprocess.Popen(["/bin/sleep", "10"]).pid'

How to reproduce

Create test project & app:

cd /tmp
django-admin.py startproject django_test
cd django_test
./manage.py startapp subprocess_test

Replace urls.py & subprocess_test/views.py with:

• urls.py:

  from django.conf.urls.defaults import *

  urlpatterns = patterns('',
    (r'^hello$', 'subprocess_test.views.hello'),
    (r'^start$', 'subprocess_test.views.start'),
  )

• subprocess_test/views.py

  from django.http import HttpResponse

  import subprocess

  def hello(request):
    return HttpResponse('Hello world!')

  def start(request):
    subprocess.Popen(["/bin/sleep", "10"])
    return HttpResponse('start done')

Test it:

./manage.py runserver 0.0.0.0:8000

Go to http://127.0.0.1:8000/hello and http://127.0.0.1:8000/start

Test result

"start" takes 10s to load and "hello" can be loaded during that time. For example, I get such a log:

[01/Feb/2011 07:20:57] "GET /hello HTTP/1.1" 200 12
[01/Feb/2011 07:21:01] "GET /start HTTP/1.1" 200 10
[01/Feb/2011 07:21:01] "GET /hello HTTP/1.1" 200 12
[01/Feb/2011 07:21:02] "GET /hello HTTP/1.1" 200 12

Using wget:

wget http://127.0.0.1:8000/start
--2011-02-01 14:31:11-- http://127.0.0.1:8000/start
Connecting to 127.0.0.1:8000... connected.
HTTP request sent, awaiting response... 200 OK
Length: unspecified [text/html]
Saving to: `start'

[           <=>                           ] 10          --.-K/s   in 9,5s    

2011-02-01 14:31:21 (1,05 B/s) - « start » saved [10]

解决方案

It looks like you don't care what the result of the system call is, so I would assume you are trying to do some sort of offline(or background) processing.

I would suggest a cleaner way of going about it rather than directly executing a program. Use a queueing system such as Gearman to queue up processing tasks and then have a separate worker that consumes items from the queue.

This has the advantage of protecting your server when large traffic spikes happen, so you don't fork off a process each time a request to that view is made. You can consume items as slow or as fast as you decide, independent of traffic.

Traffic may not be an issue, but I personally think it is a cleaner design decision as well.

这篇关于子进程阻止Django视图的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！