Gunicorn工人定期崩溃： '插座未注册'

Question

每隔一段时间（一次几个小时）gunicorn工人失败，出现以下错误：

[2014-10-29 10:21:54 +0000] [4902] [INFO] Booting worker with pid: 4902 
[2014-10-29 13:15:24 +0000] [4902] [ERROR] Exception in worker process: 
Traceback (most recent call last): 
    File "/opt/test/env/local/lib/python2.7/site-packages/gunicorn/arbiter.py", line 507, in spawn_worker 
    worker.init_process() 
    File "/opt/test/env/local/lib/python2.7/site-packages/gunicorn/workers/gthread.py", line 109, in init_process 
    super(ThreadWorker, self).init_process() 
    File "/opt/test/env/local/lib/python2.7/site-packages/gunicorn/workers/base.py", line 120, in init_process 
    self.run() 
    File "/opt/test/env/local/lib/python2.7/site-packages/gunicorn/workers/gthread.py", line 177, in run 
    self.murder_keepalived() 
    File "/opt/test/env/local/lib/python2.7/site-packages/gunicorn/workers/gthread.py", line 149, in murder_keepalived 
    self.poller.unregister(conn.sock) 
    File "/opt/test/env/local/lib/python2.7/site-packages/trollius/selectors.py", line 408, in unregister 
    key = super(EpollSelector, self).unregister(fileobj) 
    File "/opt/test/env/local/lib/python2.7/site-packages/trollius/selectors.py", line 243, in unregister 
    raise KeyError("{0!r} is not registered".format(fileobj)) 
KeyError: '<socket._socketobject object at 0x7f823f454d70> is not registered' 
... 
... 
[2014-10-29 13:15:24 +0000] [4902] [INFO] Worker exiting (pid: 4902) 
[2014-10-29 13:15:24 +0000] [5809] [INFO] Booting worker with pid: 5809 
... 

配置：

bind = '0.0.0.0:80' 
workers = 1 
threads = 4 
debug = True 
reload = True 
daemon = True 

我使用：

Python 2.7.6 
gunicorn==19.1.1 
trollius==1.0.2 
futures==2.2.0 

任何想法可能是什么原因，以及如何解决这个问题？

谢谢！

Answer 1

我面临类似的问题，我得到了从gunicorn工作人员的时间错误。我正在使用同步工作者，并且有timeout和keepalive的默认设置。 在我的使用案例中，我的http请求需要很长时间才能完成，因此同步工作人员超时。我使用curl作为发送HTTP-1.1请求的http客户端。我将超时时间增加到了一个疯狂的高数值3600即1小时，这是有效的。然而，在服务器错误日志中，我看到了和你一样的错误。这是我对这个错误的假设。 由于默认情况下，所有http 1.1请求都是持久性服务器，因此尝试通过将其重新放回队列但不超过keepalive超时重新使用连接。因此，当keepalive超时发生时，它将注销套接字，以便它不能被重用并关闭它。现在，由于我的超时值非常高，服务器尝试多次注销一个已注销的套接字，但keepalive仍然默认为5秒，因此出错。因此，我增加了“Keepalive value as well to 3600”。到目前为止它工作。

# http://gunicorn-docs.readthedocs.org/en/latest/settings.html 
timeout = 3600 # one hour timeout for long running jobs 
keepalive = 3600 

Answer 2

我报了这个错误gunicorn大约一年前，和修复应该在gunicorn 19.6.0及更高版本：https://github.com/benoitc/gunicorn/issues/1258