`google.cloud.bigquery.job.QueryJob.exception` has wrong behavior

[alexandreyc]

Hello,

According to the documentation of google.cloud.bigquery.job.QueryJob.exception, this method should returns an exception (and not raises) if something went wrong. But according to my tests this is not always the case:

from google.cloud import bigquery

client = bigquery.Client()

# Correct behavior: returns an exception and does not raise.
query = 'WRONG QUERY'
job = client.query(query)
exc = job.exception()

# Wrong behavior: raises an exception instead of returning it.
query = 'ASSERT (SELECT 2 > 3) AS "Error !!!"'
job = client.query(query)
try:
    job.exception()
except:
    print('raised!')

Am I misunderstanding the documentation or is this the correct behavior?

I'm using google-cloud-bigquery==2.6.1 with Python 3.8.6. Let me know if you need more informations.

Thanks for your help,

[HemangChothani]

@tswast @busunkim96 Please suggest.
I think something is missing for the scenario when query.state is RUNNING and library wants to retry it, it internally call the client._get_query_results method which throws an error of BadRequest.

I am not sure about to handle this in bigquery library and set the google.api_core.future.polling.PollingFuture.set_exception from the query.py file as error raised from google-cloud-core http client.

[alexandreyc]

Working on something related, I discovered that the problem does not exist when doing a double assertion:

# Correct behavior: returns an exception and does not raise.
query = 'ASSERT (SELECT 2 > 3) AS "Error !!!"; ASSERT (SELECT 2 > 3) AS "Error again !!!";'
job = client.query(query)
exc = job.exception()

Hope it can help...

Thanks,

[HemangChothani]

@tswast Backend consider "INVALID_ARGUMENT" and gave the reason "invalidQuery" when query contains ASSERT, but when query contains only one ASSERT like query = 'ASSERT (SELECT 2 > 3) AS "Error !!!"' it returns the 400 status code , hence exception raised from http client, but when query contains multiple ASSERT like 'ASSERT (SELECT 2 > 3) AS "Error !!!"; ASSERT (SELECT 2 > 3) AS "Error again !!!";', it returns 200 status code and the error result with message.

[tswast]

Regarding #451 (comment) I believe this is a difference in backend API behavior of getQueryResults between scripting jobs and non-scripting query jobs.

CC @shollyman

[tswast]

Regarding this issue, it appears that this library may be using the google-api-core library slightly incorrectly. The base class (where exception() is defined) assumes that _blocking_poll doesn't raise any exception

https://github.com/googleapis/python-api-core/blob/b51b7f587042fe9340371c1b5c8e9adf8001c43a/google/api_core/future/polling.py#L129-L134

Instead, it is expected that set_exception() is called.

We do this here:

python-bigquery/google/cloud/bigquery/job/base.py

Line 582 in 2849e56

self.set_exception(exception)

But we are missing the case when the getQueryResults backend API throws an error.

We probably need a try-catch block in QueryJob.done() as it is not expected (or documented) to raise an exception.

python-bigquery/google/cloud/bigquery/job/query.py

Line 1023 in 2849e56

self._query_results = self._client._get_query_results(

If _get_query_results does raise an exception, we might have to inspect it a bit to determine if the job is complete (failed) or it was just a transient issue.

[plamut]

@tswast Has this proposal been confirmed? If yes, I can start implementing it (as there's no assignee on the ticket yet).