ubuntu-image haschangelog tests fail due to github API rate limit exceeded

[Łukasz Zemczak]

In some recent autopkgtest runs for github PRs haschangelog test failures can be noticed that aren't actually related to real failures:

Traceback (most recent call last):
  File "/tmp/autopkgtest.SR9yz2/build.MWN/ubuntu-image/debian/tests/haschangelog", line 31, in <module>
    pull = remote.get_pull(pr_number)
  File "/usr/lib/python3/dist-packages/github/Repository.py", line 1655, in get_pull
    self.url + "/pulls/" + str(number)
  File "/usr/lib/python3/dist-packages/github/Requester.py", line 171, in requestJsonAndCheck
    return self.__check(*self.requestJson(verb, url, parameters, headers, input, cnx))
  File "/usr/lib/python3/dist-packages/github/Requester.py", line 179, in __check
    raise self.__createException(status, responseHeaders, output)
github.GithubException.GithubException: 403 {'documentation_url': 'https://developer.github.com/v3/#rate-limiting', 'message': "API rate limit exceeded for 91.189.89.216. (But here's the good news: Authenticated requests get a higher rate limit. Check out the documentation for more details.)"}

It seems like the autopkgtest infra is doing too many API calls in the selected time frame which in turn causes the test to fail. From our POV we're actually only doing two real github API calls, so probably there are some other tests that also use the same thing. One possible action would be to make the test-case wait and retry in such a case, but this would mean at the worst case the test to take over an hour.
Another solution would be to check if we can bump the rate limit (authenticating somehow?).

[Łukasz Zemczak (sil2100)]

I have prepared a workaround - it looks like just using the https API pages directly doesn't trigger API limits - at least that's what I noticed so far.

https://github.com/CanonicalLtd/ubuntu-image/pull/152

We can always re-open if this seems like not the right workaround.