BigQuery: no public method to retry/resubmit an existing job

[yan-hic]

There is currently no public method to resubmit an existing job for instance when a transient error
like 403 rateLimitExceeded is returned.

As a workaround, we currently use the following, which works for query and load :

bq = bigquery.Client()    
job_ref = bq.get_job('jobid_that_failed_because_of_ratelimit')._build_resource()
job_ref['jobReference']['jobId'] = str(uuid4())

# clear properties set by job response
 if 'query' in job_ref['configuration']:
             del (job_ref['configuration']['query']['destinationTable'])

  new_job = bq.job_from_resource(job_ref)
  new_job._begin()

Request is to have a method that works for any job type and is included in core package - as above private methods may change

[tseaver]

As luck with have it, PR #5552 (merged yesterday) moves the DEFAULT_RETRY policy into a new, public module: it does indeed retry rateLimitExceeded errors, and can be customized by users, e.g.:

from google.cloud.bigquery import retry

my_retry = retry.DEFAULT_RETRY.with_deadline(30)

[yan-hic]

Not sure the PR helps. The retry is for inserting a job, not for resubmitting an existing job.
As per Tim: https://stackoverflow.com/questions/49926546/retry-an-update-in-bigquery-does-not-seem-to-work

[dillonjohnson]

We have been retrying existing jobs previously by taking a job config and resubmitting it with a new job id. This seemed to work previously, but we are now experiencing a new issue.

The _begin() method makes a call to _to_api_repr() which is now Abstract and raises an error.

    def to_api_repr(self):
        """Generate a resource for the job."""
        raise NotImplementedError("Abstract")

    def _begin(self, client=None, retry=DEFAULT_RETRY):
        """API call:  begin the job via a POST request
        See
        https://cloud.google.com/bigquery/docs/reference/rest/v2/jobs/insert
        :type client: :class:`~google.cloud.bigquery.client.Client` or
                      ``NoneType``
        :param client: the client to use.  If not passed, falls back to the
                       ``client`` stored on the current dataset.
        :type retry: :class:`google.api_core.retry.Retry`
        :param retry: (Optional) How to retry the RPC.
        :raises: :exc:`ValueError` if the job has already begin.
        """
        if self.state is not None:
            raise ValueError("Job already begun.")

        client = self._require_client(client)
        path = "/projects/%s/jobs" % (self.project,)

        # jobs.insert is idempotent because we ensure that every new
        # job has an ID.
        api_response = client._call_api(
            retry, method="POST", path=path, data=self.to_api_repr()
        )
        self._set_properties(api_response)

If this is expected behavior, how should this be circumvented?

[tseaver]

@dillonjohnson Rather than _AsyncJob, you need to instantiate one of the concrete job classes (e.g., CopyJob, QueryJob, etc.)

[yan-hic]

No much traction but I wonder if an optional arg resubmit_on_quota_hit in the result() function would not be ideal altogether.
Downside (is it ?) is that it would change the underlying job properties like jobId

[tswast]

Closing in favor of https://github.com/googleapis/google-cloud-python/issues/9781