Reloading mod_wsgi processes safely in production

Question

Is there a way to safely restart mod_wsgi processes on demand, possibly running on multiple machines? I'm serving a django app with mod_wsgi in daemon mode with apache:

WSGIDaemonProcess myapp user=user group=user processes=30 threads=1
WSGIProcessGroup myapp

I read the suggestions in the mod_wsgi documentation but none of the suggested solutions works for me:

1. Running a monitoring script is not recommended in a production system.
2. Sending kill signals is not safe in my case. I want to avoid interrupting running requests. The reload must not bring the service down. Slowing down is fine.
3. Touching the mod_wsgi configuration file does the trick safely, but how to do it on multiple machines?

Ideally we should be able to restart the processes with one action. Any suggestions?

Note: I have to run multiple processes with one thread. My application is not thread safe.

---

Edit, I can do it from Puppet if that is the only option I have. but I prefer if there a way to do it from a special protected view. I think this is trivial to do on one machine but not multiple ones

---

Thanks a lot

Answer 1

Touching the mod_wsgi configuration file does the trick safely, but how to do it on multiple machines?

This would be solved by using a shared codebase such as from an NFS mount.

Alternatively it should be part of your version control / deployment process.

• Do you use version control currently?

  If so, create a process that ensures the file is updated. You could automate it with a post-commit hook for example.

• How do you deploy your code?

  Consider using something like Capistrano or Puppet and include an action in your deployment recipe to update that file.

Answer 2

There is no difference between (2) and (3) in respect of how existing requests are effected. Sending a kill SIGINT does not just exit the process but triggers the same sort of orderly shutdown as (2) does. In fact, (2) internally just sends a SIGINT to itself. The important thing is identifying the processes which need SIGINT sent to them. For this you should use 'display-name' option to WSGIDaemonProcess so they are named in 'ps' output and easier to identify.

Answer 3

Touching the mod_wsgi configuration file does the trick safely, but how to do it on multiple machines?

I guess this resolves the problem of how to trigger mod_wsgi to reload the scripts in a very elegant way. For the distribution of this action among several hosts at the same time, I recommend using an utility like pssh (parallel ssh). Its configuration should not be a problem.