重新加载Django网站而没有500个服务器错误？

Question

I have a Django/Gunicorn/Nginx website which I build using an Ansible playbook. I'm currently in the process of tagging those tasks in my playbook that need to run whenever I add a new feature or fix a bug and want to reload my Django code. My goal is to be able to push any such changes to production as soon as I've tested them on my staging server without having to take my production server down.

My playbook is currently set up so that whenever I want to push changes to production, it will delete the entire website directory and rebuild the Pip virtual environment so that there's no leftover "cruft". The problem is that when I run the Ansible playbook, I get a momentary 500 server error at the point at which it deletes the website directory. Then, as soon as the next task that checks out the code from Github runs, the error goes away. Clearly, whenever I reload my code base, if any user is hitting my website at the moment that task runs, they'll get this 500 error too. I've been thinking that once Gunicorn is started and my Django code is loaded into memory, I can delete and reload my code and restart Gunicorn without any interruption to service. This other Stackover question seems to that should be the case. However, my tests indicate otherwise.

When a client browser sends a request to Django, does Django reload all the .pyc files into memory? Is that what's causing this problem? Is there any way to prevent this momentary outage when I reload my Django code without removing the task that deletes the old code?

Here are excerpts from my Ansible playbook:

- name: configure web server
  hosts: webservers
  gather_facts: True
  tasks:
     # This task causes momentary 500 server error
     - name: delete any existing project repo
       file: >
         path={{ repo_path }}
         state=absent
       tags: reload

    - name: check out {{ repo_version }} of Github repo
      git: >
        repo={{ repo_url }}
        version={{ repo_version }}
        dest={{ repo_path }}
        accept_hostkey=yes
      register: checkout
      until: checkout|success
      retries: 5
      delay: 10
      become: true
      become_user: "{{ me }}"
      tags: reload

    # ... more tasks

    - name: install python packages into virtual environment
      pip: >
        requirements={{ repo_path }}/requirements/{{ server_tier }}.txt
        state=present
        virtualenv={{ venvs_path }}/{{ commit_hash }}
      tags: reload

    # ... more tasks

    - name: reload gunicorn
      command: pkill -HUP gunicorn
  become: true
  become_user: root
  tags: reload

Answer 1

Solution : Make your django root directory a symb link.

Let me explain my solution in details. Make sure that your code root is a symbolic link to actual version of code that is serving the traffic. During a deployment your ansible script will essentially create a new virtualenv, and install the requirements. And pull code in a new directory which should have current timestamp in the name. And you do whatever you need to do make it ready to serve traffic. Then in the last step you just start pointing the django root directory link to new location.

/home/ubuntu/project/current -> /home/ubuntu/project/releases/145678/
/home/ubuntu/project/releases/123456/
/home/ubuntu/project/releases/134567/
/home/ubuntu/project/releases/145678/

In the above directory structure, you can point current to one of the releases. With every new deployment you can create a new release and then subsequently point current to new release directory.

After this you can restart your gunicorn to refresh the settings from new codebase.

You can have a cron job to remove very old releases from the machine.

Let me know if you need more clarification.