`Client.query()` sometimes raises 404, despite the query job being successfully started #2134

tswast · 2025-02-18T20:05:43Z

Environment details

OS type and version: Google Colab
Python version: python --version 3.11
pip version: pip --version N/A
google-cloud-bigquery version: pip show google-cloud-bigquery 3.25.0

Steps to reproduce

It's a bit difficult to reproduce, but it seems to happen most often with creating a vector index via CREATE VECTOR INDEX DDL as generated by bigframes.bigquery.create_vector_index.

Code example

Long Python sample, sorry

PROJECT_ID = "my-project"  # @param {type:"string"}
DATASET_ID = "my_dataset"  # @param {type:"string"}
CONNECTION_ID = "bqml_llm_conn"  # @param {type:"string"}

# Set the project id
! gcloud config set project {PROJECT_ID}

REGION = "US"  # @param {type: "string"}

## Text embedding variables

TEXT_TABLE_ID = "PATENT_TABLE_BF"
TEXT_MODEL_ID = "llm_embedding_model"
TEXT_EMBEDDING_TABLE_ID = "patent_embedding_BF_test"
TEXT_EMBEDDING_TABLE_ID2 = "patent_embedding"
TEXT_MODEL_ENDPOINT = "text-embedding-004"
TEXT_VECTOR_INDEX_ID = "patent_index_BF"

import bigframes.pandas as bf
import bigframes.ml as bf_ml
import bigframes.bigquery as bf_bq
import bigframes.ml.llm as bf_llm


from google.cloud import bigquery
from google.cloud import storage

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

import pandas as pd
from IPython.display import Image, display
from PIL import Image as PILImage
import io

bf.options.bigquery.project = PROJECT_ID
bf.options.bigquery.location = REGION
bf.options.bigquery.ordering_mode = "partial"

text_model = bf_llm.TextEmbeddingGenerator() ## which endpoint this uses??

publications = bf.read_gbq('patents-public-data.google_patents_research.publications')
keep = (publications.embedding_v1.str.len() > 0) & (publications.title.str.len() > 0) & (publications.abstract.str.len() > 0)
publications = publications[["publication_number", "title", "abstract"]].rename(columns={'abstract': 'content'})

## creating embeddings for 10000 rows only for testing, change in the line below for embedding higher number of rows
publications_subset = publications[publications["content"].str.len() > 30].peek(10000)
## publications_subset = publications_subset[["publication_number", "title", "abstract"]].rename(columns={'abstract': 'content'})


embedding = text_model.predict(publications_subset)[["publication_number", "title", "content", "ml_generate_embedding_result","ml_generate_embedding_status"]]

## filter out rows where the embedding generation failed. the embedding status value is empty if the embedding generation was successful
embedding = embedding[~embedding["ml_generate_embedding_status"].isnull()]

# store embeddings in a BQ table
embedding.to_gbq(f"{DATASET_ID}.{TEXT_EMBEDDING_TABLE_ID}", if_exists='replace')

## python code to create index
bf_bq.create_vector_index(
    table_id = f"{DATASET_ID}.{TEXT_EMBEDDING_TABLE_ID}",
    column_name = "ml_generate_embedding_result",
    replace= True,
    index_name = "bf_python_index",
    distance_type="cosine",
    index_type= "ivf"
)

Unfortunately, even this is not a guarantee, but it does seem pretty reproducible on one of the projects my colleague is using.

Stack trace

/usr/local/lib/python3.10/dist-packages/bigframes/session/_io/bigquery/__init__.py in start_query_with_client(bq_client, sql, job_config, max_results, timeout, api_name, metrics)
    229 
    230     try:
--> 231         query_job = bq_client.query(sql, job_config=job_config, timeout=timeout)
    232     except google.api_core.exceptions.Forbidden as ex:
    233         if "Drive credentials" in ex.message:

/usr/local/lib/python3.10/dist-packages/google/cloud/bigquery/client.py in query(self, query, job_config, job_id, job_id_prefix, location, project, retry, timeout, job_retry, api_method)
   3490             )
   3491         elif api_method == enums.QueryApiMethod.INSERT:
-> 3492             return _job_helpers.query_jobs_insert(
   3493                 self,
   3494                 query,

/usr/local/lib/python3.10/dist-packages/google/cloud/bigquery/_job_helpers.py in query_jobs_insert(client, query, job_config, job_id, job_id_prefix, location, project, retry, timeout, job_retry)
    157             return query_job
    158 
--> 159     future = do_query()
    160     # The future might be in a failed state now, but if it's
    161     # unrecoverable, we'll find out when we ask for it's result, at which

/usr/local/lib/python3.10/dist-packages/google/cloud/bigquery/_job_helpers.py in do_query()
    143 
    144             try:
--> 145                 query_job = client.get_job(
    146                     job_id,
    147                     project=project,

/usr/local/lib/python3.10/dist-packages/google/cloud/bigquery/client.py in get_job(self, job_id, project, location, retry, timeout)
   2199         span_attributes = {"path": path, "job_id": job_id, "location": location}
   2200 
-> 2201         resource = self._call_api(
   2202             retry,
   2203             span_name="BigQuery.getJob",

/usr/local/lib/python3.10/dist-packages/google/cloud/bigquery/client.py in _call_api(self, retry, span_name, span_attributes, job_ref, headers, **kwargs)
    831                 name=span_name, attributes=span_attributes, client=self, job_ref=job_ref
    832             ):
--> 833                 return call()
    834 
    835         return call()

/usr/local/lib/python3.10/dist-packages/google/api_core/retry/retry_unary.py in retry_wrapped_func(*args, **kwargs)
    291                 self._initial, self._maximum, multiplier=self._multiplier
    292             )
--> 293             return retry_target(
    294                 target,
    295                 self._predicate,

/usr/local/lib/python3.10/dist-packages/google/api_core/retry/retry_unary.py in retry_target(target, predicate, sleep_generator, timeout, on_error, exception_factory, **kwargs)
    151         except Exception as exc:
    152             # defer to shared logic for handling errors
--> 153             _retry_error_helper(
    154                 exc,
    155                 deadline,

/usr/local/lib/python3.10/dist-packages/google/api_core/retry/retry_base.py in _retry_error_helper(exc, deadline, next_sleep, error_list, predicate_fn, on_error_fn, exc_factory_fn, original_timeout)
    210             original_timeout,
    211         )
--> 212         raise final_exc from source_exc
    213     if on_error_fn is not None:
    214         on_error_fn(exc)

/usr/local/lib/python3.10/dist-packages/google/api_core/retry/retry_unary.py in retry_target(target, predicate, sleep_generator, timeout, on_error, exception_factory, **kwargs)
    142     for sleep in sleep_generator:
    143         try:
--> 144             result = target()
    145             if inspect.isawaitable(result):
    146                 warnings.warn(_ASYNC_RETRY_WARNING)

/usr/local/lib/python3.10/dist-packages/google/cloud/_http/__init__.py in api_request(self, method, path, query_params, data, content_type, headers, api_base_url, api_version, expect_json, _target_object, timeout, extra_api_info)
    492 
    493         if not 200 <= response.status_code < 300:
--> 494             raise exceptions.from_http_response(response)
    495 
    496         if expect_json and response.content:

NotFound: 404 GET https://bigquery.googleapis.com/bigquery/v2/projects/my-project/jobs/c540a00e-05ed-4caf-b76c-b23d16b9617f?projection=full&location=US&prettyPrint=false: Not found: Job my-project:US.c540a00e-05ed-4caf-b76c-b23d16b9617f

Note that it is getting a 404 here:

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

Line 145 in b03a2af

query_job = client.get_job(

This is after a Conflict exception, meaning the job was created successfully, the retry failed because it already exists, but then getting that already existing job failed with 404. Something funky is going on with the networking I think for this case to happen, but it should be possible to simulate in unit tests, at least.

The text was updated successfully, but these errors were encountered:

tswast · 2025-02-18T20:06:19Z

I can take this one since it's blocking some bigframes work.

tswast · 2025-02-18T21:26:28Z

One thing to note is that I asked my colleague: "Does a job with that ID show up in the BQ UI?" Answer: "no it does not"

This implies to me that we should also retry the query job with a new ID in some of these cases (despite the Conflict) error.

tswast self-assigned this Feb 18, 2025

product-auto-label bot added the api: bigquery Issues related to the googleapis/python-bigquery API. label Feb 18, 2025

tswast added type: bug Error or flaw in code with unintended results or allowing sub-optimal usage patterns. priority: p1 Important issue which blocks shipping the next release. Will be fixed prior to next release. labels Feb 18, 2025

tswast linked a pull request Feb 18, 2025 that will close this issue

fix: retry 404 errors in Client.query(...) #2135

Open

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`Client.query()` sometimes raises 404, despite the query job being successfully started #2134

`Client.query()` sometimes raises 404, despite the query job being successfully started #2134

tswast commented Feb 18, 2025

tswast commented Feb 18, 2025

tswast commented Feb 18, 2025

Client.query() sometimes raises 404, despite the query job being successfully started #2134

Client.query() sometimes raises 404, despite the query job being successfully started #2134

Comments

tswast commented Feb 18, 2025

Environment details

Steps to reproduce

Code example

Stack trace

tswast commented Feb 18, 2025

tswast commented Feb 18, 2025

`Client.query()` sometimes raises 404, despite the query job being successfully started #2134

`Client.query()` sometimes raises 404, despite the query job being successfully started #2134