The Wizard: Side effects
See the previous episode here.
He has questions. Nevertheless, he has not yet arrived. And you know you have answers. You interrupt your summoning, and sit down. You can continue in another moment. Everything happens for a reason. You just hope the reason is good.
N appears right next to you.
- Hey, what’s up?
- I'm setting up the auto-scaling configuration of the new project. Nothing huge.
- Cool. I need some help.
He has progressed immensely since the last time you saw him. People are starting to ask him for help now. You can’t help but feel pride when you see what he has become.
- Yeah sure, show me.
- Remember that nginx role you helped me create?
- What about it?
- It’s flawed.
Interesting. He did realize after all.
- Is it, now?
- Yes. I usually run my Ansible playbooks every night, in order to ensure that all my infrastructure and services are at a certain state. At least the ones in charge of server and middleware configuration.
That’s good stuff, you think.
- That’s good stuff.
- It is. It works as long as my playbooks are idempotent.
- Are your playbooks idempotent?
- Yes, except for the ones I use for application deployment.
- What’s your issue then?
- My nginx role. The service stopped unexpectedly, and my role wouldn’t restart it.
He realized the hard way. That is probably a good thing. It is the best way to learn.
- Okay. Let’s take a look.
You summon the enclosure you both worked on on that occasion. You need it in order to properly analyze the spell. You check the structure of it:
# tasks/main.yml --- - name: Install epel-release yum: name: epel-release state: present - name: Install nginx yum: name: nginx state: present notify: Start nginx
It all comes back to you. You remember noticing the little issue with this proposal. It was a time bomb. But then again, he found it. It’s precisely what you wanted.
- Well, you’re only starting your service when installing nginx.
- So if your service is stopped, and you launch your playbook again, nothing will change when nginx is already installed.
- Yes. So this whole testing thing… I don’t think it’s a good idea. I refactored my code, the tests passed, and the service is not running in production because of a side effect.
Patience is key.
- Did you test that side effect?
- Well… no. How could I? It’s a side effect.
- Write the code that would cause your side effect, and then see if your actual code is able to handle that scenario.
He considers the idea for half a second.
- So I’m supposed to put code which makes my components fail inside of my role? That doesn’t make any sense, does it?
- You don’t put it inside your role. You put it alongside your code. Watch.
You reflect for a second, structuring your thoughts, trying to imagine exactly what you need. You analyze the dimensions of the enclosure. Depth, width and height. It won’t suffice for what you have in mind. Once you figure it out, you close your eyes, and expand the enclosure in order to create a new dimension, invisible to the naked eye. Inside this new pocket within time and space, you place the side effect of the spell, just as he described it:
# molecule/default/side_effect.yml --- - name: Stop nginx hosts: all tasks: - name: Stop nginx service: name: nginx state: stopped
This will probably suffice, you say to yourself.
- Would you say this code is capable of recreating the side effect you described earlier?
- Yes, I would.
- Great, let’s use it now.
You concentrate as you modify the structure and logic of the enclosure in order to use the new dimension:
# molecule/default/molecule.yml scenario: name: default test_sequence: - destroy - syntax - lint - create - converge - side_effect - converge - idempotence - verify
- So we’re converging twice?
- Yes. You need to test whether your code is able to correct the state of your component after a side effect.
- Yes, indeed.
- Now, launch your test.
N claps his hands, and as he separates them, a stream of bright red light appears between them:
=================================== FAILURES =================================== __________________ test_nginx_is_running[ansible://instance] ___________________ host = <testinfra.host.Host object at 0x108390290> def test_nginx_is_running(host): # Given nginx = host.service('nginx') # Then > assert nginx.is_running E assert False E + where False = <service nginx>.is_running tests/test_default.py:20: AssertionError
He looks stunned for a while.
- So we weren’t testing the whole way, were we?
- No, we weren’t. I mean, we were not testing for all of the edge cases we could have had. You can’t possibly test for all of them. There’s way too many. This was an interesting one though.
- I see.
- Well, you’re in red again. Go to green.
N modifies the spell structure.
# tasks/main.yml --- # tasks file for nginx-tdd - name: Install epel-release yum: name: epel-release state: present - name: Install nginx yum: name: nginx state: present - name: Start nginx service: name: nginx state: started enabled: true
It looks good.
- And now let me test…
N does exactly the same thing he did before. Only this time, bright green light emanates from his palms:
============================= test session starts ============================== platform darwin -- Python 2.7.11, pytest-3.3.2, py-1.5.2, pluggy-0.6.0 rootdir: /Users/sebiwi/stuff/ocac/skool-ops/craftmanship-ops/ansible-tdd/ansible-nginx-tdd/molecule/default, inifile: plugins: testinfra-1.7.1 collected 3 items tests/test_default.py ... [100%] =========================== 3 passed in 7.78 seconds =========================== Verifier completed successfully.
N looks at you.
- That was simple.
- Was it?
- I mean, I could have made my tests better from the beginning.
- One bug in production is just a thing you didn’t test.
- It seems like overkill for a simple nginx installation though.
He is right. But he needs to see the bigger picture.
- It probably is. But now you know that you’re really testing interesting scenarios, the ones you really want. Besides, think about it. The possibilities are endless.
- What do you mean?
- You only have a single node component here, but you can use the same principle in order to test incredibly complex things.
- Such as?
- High availability on distributed components, leader election, node failover… you name it. You can simulate a disaster on a side effect and see how your code will react to it.
He reflects. He’s starting to get it.
- Can I help you with anything else?
Your attention already shifted back to the battlefield. What now?
- No, I think that will do. I’ll probably go add this to my other playbooks. You’re right, the possibilities are…
You’re no longer there again. N looks around in confusion.
- How does he do that ?