Auto-Terminating Long-Running Requests in Django

Question

I've got a setup with Apache + mod_wsgi running django code, and I'd like to add a layer of protection in case a non-terminating view slips in. Something that kills-off a requests exceeding, say, 30 seconds would be ideal.

For testing I've just put a time.sleep(60) in a view.

I've tried the TimeOut 30 setting in Apache, but I still get curl returning after 60 seconds.

I see mod_wsgi itself offers three different time-out values, but none of them seem to apply to a long-running request.

Is there a standard piece of Django middleware for this or is there a knob I'm missing on Apache or mod_wsgi?

Answer 1

It is actually really difficult to terminate a single Python request thread in a multithread application. The best you can do is make a decision to shutdown the whole process and restart it. Because such an action will disrupt concurrent requests, you would as a result really need to restrict yourself to a single threaded configuration.

Even then the support in mod_wsgi 3.X isn't ideal for this. There is inactivity-timeout for daemon mode, but it actually causes process to be restarted in two situations. The first is when there are no requests at all and process is idle. The second is when all request threads have blocked and timeout expires.

In mod_wsgi 4.X (in repository trunk at this time), the two concepts have been separated and now inactivity-timeout only applies to completely idle process with no concurrent requests. A new blocked-timeout has been added to separately specific a timeout for when whole process is blocked. It is this latter one you can use.

If you want to learn more about the new option, you will need to go over to the mod_wsgi mailing list to discusss it.