'_AsyncQuery.cancel' fails to update from returned resource

[tseaver]

E.g.:

Traceback (most recent call last):
  File ...
    job.cancel()
  File ".../google/cloud/bigquery/job.py", line 378, in cancel
    self._set_properties(api_response)
  File ".../google/cloud/bigquery/job.py", line 262, in _set_properties
    self._scrub_local_properties(cleaned)
  File ".../google/cloud/bigquery/job.py", line 1050, in _scrub_local_properties
    configuration = cleaned['configuration']['query']
KeyError: 'configuration'

The docs for job.cancel show that the job resource is in a job subkey of the response.

[dhermes]

@tseaver Germane to this change, I've run the following before to get a sense of our actual usage of streaming:

nosetests system_tests/storage.py system_tests/bigquery.py \
  --with-coverage  \
  --cover-branches  \
  --cover-tests  \
  --cover-package=gcloud.streaming  \
  --cover-package=gcloud.bigquery  \
  --cover-package=gcloud.storage  \
  --nocapture

Could be worthwhile to do it and see what code paths we aren't testing against the backend?

(NOTE: I'm aware this is using the old gcloud package and nosetests, sorry I am in a hurry and didn't update for the py.test equivalent.)

[tseaver]

@dhermes I don't think that is germane to this issue, which doesn't mess with streaming at all. Maybe you meant #1431?

[dhermes]

No I just meant that if you got a sense for lines uncovered during system tests then an error like this would be detectable. The streaming bit was just an example so you could do the same locally.

Essentially this error is due to the fact that our code and unit tests make a bad assumption about the payload.

[tseaver]

OK. We don't have any system test which exercises _AsyncJob.cancel(): I found the problem while reproducing #2368.

I'm not convinced that we need to get to 100% coverage for system tests, but we should at least have "garden path" coverage for all the API requests.

[dhermes]

Ha. Never heard "garden path" before! I agree having full coverage in system tests isn't necessary (or possible, since error cases may never occur), just wanted to share a script I had used before