BigQuery: client.create_table raises Conflict instead of AlreadyExists

[jpuig-mind]

Environment details

1. API: BigQuery

2. OS type and version:

   $ lsb_release -a
   No LSB modules are available.
   Distributor ID: Ubuntu
   Description:    Ubuntu 20.04 LTS
   Release:        20.04
   Codename:       focal

3. Python version and virtual environment information:

   $ python3 --version
   Python 3.8.2

4. google-cloud-bigquery version:

   $ pip show google-cloud-bigquery
   Name: google-cloud-bigquery
   Version: 1.24.0
   Summary: Google BigQuery API client library
   Home-page: https://github.com/GoogleCloudPlatform/google-cloud-python
   Author: Google LLC
   Author-email: googleapis-packages@google.com
   License: Apache 2.0
   Location: /home/REDACTED/.local/lib/python3.8/site-packages
   Requires: protobuf, google-api-core, google-cloud-core, google-auth, google-resumable-media, six
   Required-by:

Steps to reproduce

1. Have or create a table in BigQuery
2. Try to create it again using client.create_table()
3. Confirm that the raised expression is google.api_core.exceptions.Conflict
   • One would expect it to be google.api_core.exceptions.AlreadyExists
   • Checked client.delete_table() with a non-existent table and the result is google.api_core.exceptions.NotFound

Code example

Slightly modified version of https://cloud.google.com/bigquery/docs/tables#python

from google.cloud import bigquery
from google.api_core.exceptions import AlreadyExists, Conflict

# Construct a BigQuery client object.
client = bigquery.Client()

# TODO(developer): Set table_id to the ID of the table to create.
table_id = "your-project.your_dataset.your_table_name"

table = bigquery.Table(table_id)
table = client.create_table(table)  # Make an API request.
try:
    table = client.create_table(table)  # Make an API request. Second time.
except AlreadyExists:
    print("Caught google.api_core.exceptions.AlreadyExists")
except Conflict:
    print("Caught google.api_core.exceptions.Conflict")

table = client.create_table(table)  # Make an API request. Third time for Stack Trace

Stack trace

$ /usr/bin/python3 REDACTED/example.py
Caught google.api_core.exceptions.Conflict
Traceback (most recent call last):
  File "REDACTED/example.py", line 19, in <module>
    table = client.create_table(table)  # Make an API request. Third time for Stack Trace
  File "/home/REDACTED/.local/lib/python3.8/site-packages/google/cloud/bigquery/client.py", line 543, in create_table
    api_response = self._call_api(
  File "/home/REDACTED/.local/lib/python3.8/site-packages/google/cloud/bigquery/client.py", line 556, in _call_api
    return call()
  File "/home/REDACTED/.local/lib/python3.8/site-packages/google/api_core/retry.py", line 281, in retry_wrapped_func
    return retry_target(
  File "/home/REDACTED/.local/lib/python3.8/site-packages/google/api_core/retry.py", line 184, in retry_target
    return target()
  File "/home/REDACTED/.local/lib/python3.8/site-packages/google/cloud/_http.py", line 423, in api_request
    raise exceptions.from_http_response(response)
google.api_core.exceptions.Conflict: 409 POST https://bigquery.googleapis.com/bigquery/v2/projects/REDACTED/datasets/REDACTED/tables: Already Exists: Table REDACTED:REDACTED.your_table_name

[skelliest]

Also, there does not seem to be a way to distinguish between the two types of Conflict errors: AlreadyExists and Aborted. This is problematic because the existence of the table conditional on each error is different. AlreadyExists is supposed to have grpc_status_code set to grpc.StatusCode.ALREADY_EXISTS, but on my machine it is not.

[HemangChothani]

@jpuig-mind @skelliest Agreed with you statement, it returns http code rather than grpc_status_code because bigquery is rest client and it works over http , not on grpc so it returns http status code.

As you can see below in source code also, to handle exception use google.api_core.exceptions.Conflict.

python-bigquery/google/cloud/bigquery/client.py

Lines 549 to 552 in be86de3

	except google.api_core.exceptions.Conflict:
	if not exists_ok:
	raise
	return self.get_table(table.reference, retry=retry)

[jpuig-mind]

@HemangChothani Thanks for the clarification.

For clarity then, shouldn't the exception handling be something more like the following?

except google.api_core.exceptions.Conflict as e: 
    if not exists_ok: 
        raise google.api_core.exceptions.AlreadyExists(e.message, e.errors, e.response)
    return self.get_table(table.reference, retry=retry) 

I suggest this solution since it's the BigQuery client that should hold the "if Conflict then AlreadyExists" knowledge, and take the guesswork off the hands of the user.

I apologise if I made any mistake, python is not my first language.

Cheers!

[HemangChothani]

@jpuig-mind I think it's bit difficult to bifurcate AlreadyExists and Aborted errors, if consider your code of snippet then it raise the error from google/cloud/_http.py because status code is not 2XX.

code snippet:

    if not 200 <= response.status_code < 300:             
        raise exceptions.from_http_response(response)

and then another exception raised what you suggest:

    except google.api_core.exceptions.Conflict as e: 
        if not exists_ok: 
            raise google.api_core.exceptions.AlreadyExists(e.message, e.errors, e.response)

I don't think it's good idea, but still i would like to ask @plamut
It would be great if you give any suggestion here?

[HemangChothani]

@plamut any suggestion?

[plamut]

@HemangChothani If I understand correctly, the concern is that we first create an exception from the HTTP response (in api_core), but would then again have to transform the very same exception in BigQuery as proposed above?

Ideally, we should only translate the exception once (in api-core). However, if api-core does not have all information to pick the right exception type in all cases (e.g. exists_ok flag), then it's up to BigQuery to decide which flavor of Conflict error to actually raise.

It would probably also be beneficial to keep the link with the original exception:

except api_core.exceptions.Conflict as exc: 
    if not exists_ok:
        rebranded_error = api_core.exceptions.AlreadyExists(exc.message, exc.errors, exc.response)
        six.raise_from(rebranded_error, exc)

Or did I misunderstand the question?

[HemangChothani]

@plamut Thanks for the help,appreciate it. The main concern is how to distinguish errors between Already Exists and Aborted errors as both inherit Conflict class. I found only one option to compare an exc.message with a static string which might be not appropriate.