BigQuery: insert_rows() fails when a repeated field is missing

[danych2]

Using the method insert_rows to insert a row missing a REPEATED field returns an error. Repeated fields should be nullable by default, so it should not matter if a repeated field is not provided.
The method insert_rows_json, which insert_rows uses to make the API call seems to work fine.

Environment details

Python 2.7.15+
google-cloud-bigquery 1.21.0

Steps to reproduce

• Create a table with a repeated field
• Insert a row into the table using insert_rows, passing a dict where the repeated field key-value pair is missing.

[tseaver]

@tswast AFAIK, this would be a back-end issue.

[plamut]

Seems related to #9207.

[plamut]

I was able to reproduce it in Python 3, too. Posting a minimal reproducible example (the table must already exist):

from __future__ import print_function

from google.cloud import bigquery


PROJECT = "project_name"
DATASET = "dataset_name"
TABLE = "table_name"

client = bigquery.Client(PROJECT)
dataset_ref = client.dataset(DATASET)
table_ref = dataset_ref.table(TABLE)

schema = [
    bigquery.SchemaField("int_col", "INTEGER", mode="REQUIRED"),
    bigquery.SchemaField("repeated_col", "INTEGER", mode="REPEATED"),
])

rows = [
    # {"int_col": 5, "repeated_col": [10, 20, 30]},  # works
    # {"int_col": 8, "repeated_col": []},            # works 
    # {"int_col": 9, "repeated_col": None},          # "Field value cannot be empty."
    {"int_col": 12},                                 # "Field value cannot be empty."
]

print("Inserting rows...")
errors = client.insert_rows(table_ref, rows, selected_fields=schema)
print("Insert errors:,", errors)

The error is reproducible if the repeated field is not present in the data, or if it's set to None (even if insert_rows_json() is used directly). On the other hand, things work fine if the field is explicitly set to an empty list. All three cases should probably indeed be treated in the same way, though.

Classifying as P2, since a straightforward workaround exists.

[tswast]

Actually, we do have a client-side issue here. Inserting {"int_col": 12} works with insert_rows_json, but not insert_rows.

We need to update the logic for insert_rows to omit any missing values in the generated resource rather than the current logic, where it explicitly sets missing values to None.

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

Line 431 in b387134

record[subname] = _field_to_json(subfield, subvalue)

$ python write_python.py
----------------------------------------
insert_rows_json
----------------------------------------
Inserting populated rows...
Insert errors:

Inserting null rows...
Insert errors:
{'index': 0, 'errors': [{'reason': 'invalid', 'location': 'repeated_col', 'debugInfo': '', 'message': 'Field value cannot be empty.'}]}

Inserting missing rows...
Insert errors:

----------------------------------------
insert_rows
----------------------------------------
Inserting populated rows...
Insert errors:

Inserting null rows...
Insert errors:
{'index': 0, 'errors': [{'reason': 'invalid', 'location': 'repeated_col', 'debugInfo': '', 'message': 'Field value cannot be empty.'}]}

Inserting missing rows...
Insert errors:
{'index': 0, 'errors': [{'reason': 'invalid', 'location': 'repeated_col', 'debugInfo': '', 'message': 'Field value cannot be empty.'}]}

Source:

from google.cloud import bigquery


client = bigquery.Client()

print("-" * 40)
print("insert_rows_json")
print("-" * 40)

rows = [
    {"repeated_col": []},  # works
    {"int_col": 5, "repeated_col": [10, 20, 30]},  # works
    {"int_col": 8, "repeated_col": []},            # works
]

print("Inserting populated rows...")
errors = client.insert_rows_json("my_dataset.google_cloud_python_9602", rows)
print("Insert errors:")
for error in errors:
    print(error)
print()

rows = [
    {"int_col": 9, "repeated_col": None},          # "Field value cannot be empty."
]
print("Inserting null rows...")
errors = client.insert_rows_json("my_dataset.google_cloud_python_9602", rows)
print("Insert errors:")
for error in errors:
    print(error)
print()

rows = [
    {"int_col": 12},                                 # works
]
print("Inserting missing rows...")
errors = client.insert_rows_json("my_dataset.google_cloud_python_9602", rows)
print("Insert errors:")
for error in errors:
    print(error)
print()


print("-" * 40)
print("insert_rows")
print("-" * 40)

schema = [
    bigquery.SchemaField("int_col", "INTEGER", mode="REQUIRED"),
    bigquery.SchemaField("repeated_col", "INTEGER", mode="REPEATED"),
]

rows = [
    {"repeated_col": []},  # works
    {"int_col": 5, "repeated_col": [10, 20, 30]},  # works
    {"int_col": 8, "repeated_col": []},            # works
]

print("Inserting populated rows...")
errors = client.insert_rows("my_dataset.google_cloud_python_9602", rows, selected_fields=schema)
print("Insert errors:")
for error in errors:
    print(error)
print()

rows = [
    {"int_col": 9, "repeated_col": None},          # "Field value cannot be empty."
]
print("Inserting null rows...")
errors = client.insert_rows("my_dataset.google_cloud_python_9602", rows, selected_fields=schema)
print("Insert errors:")
for error in errors:
    print(error)
print()

rows = [
    {"int_col": 12},                                 # "Field value cannot be empty."
]
print("Inserting missing rows...")
errors = client.insert_rows("my_dataset.google_cloud_python_9602", rows, selected_fields=schema)
print("Insert errors:")
for error in errors:
    print(error)
print()

[tswast]

As I don't believe the backend has changed it's behavior for this case, any backend issue I were to file would be considered a feature request.