Swarming: Adds CountTasks endpoint & Refactors swarming Prpc request

[IvanBM18]

As part of the Swarming backpressure initiative, we needed to launch request to the prpc's swarming.v2.Task/CountTasks endpoint so we can calculate the amount of fuzz tasks needed to schedule, to achieve this i moved the prpc request logic to its own module so that we can reuse code and keep responsibilities in separate files.

Changes

• Created a new api.py file in the swarming module.
  • Moved the request logic from __init__.py to api.py
  • Removed push_swarming_task from init.py
  • Updated service.py to use SwarmingAPI directly.
  • Created a new class SwarmingAPI to handle the prpc request logic such as token/auth logic, swarming config handling, and any other request logic
• Created new unit tests for this api and Updated existing unit tests.

Tests

• Created unit tests for the new api & endpoint
• Scheduled swarming jobs to verify that the refactor didn't broke the existing swarming logic
  • See example logs

[Xeicker]

For future PRs, when you move code and then add functionality it is a good practice to do each of those steps in separate commits for an easier review

[Xeicker]

Other than some nits changes LGTM

[letitz]

IIUC this means we'll make requests with no tokens if we fail to get a token, which is bound to fail? ISTM we should bail out and not make a request if we cannot get an auth token for it.

I would either expose to callers that SwarmingApi methods might raise either GoogleAuthError or RequestException, or wrap those in a new exception type here: SwarmingError and let the caller deal with that.

Edit: I see your comment on the test, that this will show up as 401 errors. I still find it a bit surprising to ask the system to do something we know will fail, but I must admit it works. I'll let @javanlacerda decide how he feels about it.

[letitz]

I would expect that debug-level logs are not logged in production, and only enabled if requested specifically somehow (environment variable?) during e.g. local testing. Anyway, thanks for considering it!

[letitz]

LGTM % adding one last test. Outstanding comments are left for @javanlacerda's consideration :)

[letitz]

We should have a test for what happens when the API raises an exception we can catch.