How do you run these tests without using the unittest module? Are they imported somewhere into a main file which runs all the tests?

Our main test runner is pytest which scans for functions that start with test_ and handles the Python asserts in a way similar to how unittest would the assert functions, so it's pretty neat.

If you do:

pip install pytest
pytest

it will run all the tests.

In this specific repo, since there's dozens of samples ,we use nox which helps instrument different sessions etc. But really in here it will just call pytest.

while constants are good practice in production code, in sample code they add a layer of indirection that can be annoying when including a snippet in cloud site. Strongly prefer this to go near where it's used.

done

For sample code, prefer pulling out discrete values and giving them names:

results = client.query(CPU_METRIC, minutes=5)
results = list(query)
first_result = results[0]
cpu_load = first_result.points[0].value

Also, shouldn't that query have a limit of 1?

Done. API doesn't support limits on queries.

I'd recommend scale_up as a more more descriptive name. Then, you can probably drop the args section and just describe the behavior in the description.

scale_up seems misleading in the case of downscale.

And up is less misleading?

Sorry, misunderstood, thought you meant the function name. Done.

move the while to main.

I had it in main, but I moved it out here to make it easier to test main. Otherwise testing while True loops is annoying. Would probably have to make a bunch of changes just to test the code, this seemed like the simpler option.

The tests shouldn't influence flow control. We can figure out how to break out of the loop. Likely by using mock to insert a keyboard interrupt.

Any reason to have this as a separate file? For the sake of samples, it seems it would be most useful to locate this near the code that uses these strategies.

The idea is we are going to add a few alternative strategies such as exponential, and have them all organized in here.

I see. If that's valuable and matches with how this will be presented in the docs, then that's fine.

Ya doc is still a WIP so once we have a draft of that I'll take another look at this and if it's pointlessly complicating things I'll move it back into autoscaler.py.