Fetching next batch of results using BigQuery's dbapi

Currently, the maximum number of rows that you can get when iterating using fetchmany, fetchall or fetchone is limited by arraysize.

I see that those methods are using iterator returned by _query_data.fetch_data() with max_results set to arraysize
https://github.com/GoogleCloudPlatform/google-cloud-python/blob/36304f7448677cff2f89558ef87c374e904a0492/bigquery/google/cloud/bigquery/dbapi/cursor.py#L171-L172

However, shouldn't it automatically fetch the next page of results when we are finished iterating over the current batch of results? Otherwise, it makes it impossible to retrieve full results from the query without settingarraysize to an arbitrary large number.

Here is what I've tried:

cursor.execute('SELECT * FROM dataset.table')

while True:
    rows = cursor.fetchmany(5000)
    if not rows:
        break

    for row in rows:
        print(row)

cursor.execute('SELECT * FROM dataset.table')

row = cursor.fetchone()
while row:
   print(row)
   row = cursor.fetchone()

I am using google-cloud-bigquery 0.26.0

I'm not sure what you are asking. Under the covers the fetch_data method returns an iterator, so it would be possible to prefetch results, but why would you want to?

The DB-API specification says that arraysize must default to 1, so that's what the BigQuery DB-API does to stay compliant. As with any DB-API implementation, you must set the arraysize manually if you want a larger batch size.

Unless I am missing something, there is currently no way to fetch all rows for a query? The only workaround is to set arraysize to some arbitrary large number, because we don't know beforehand how many rows the result for the query will contain.

What's wrong with using fetchall?

But yes, for best performance you will need to increase the arraysize parameter before calling fetchall

fetchall should not be limited by arraysize. It's just the batch size it uses to fetch results.

This is what I thought as well, but I've tried fetchall and it just returns 1 row or what I specify for arraysize.

Ah, okay. I'll investigate further. Thanks for you patience.

I'm able to reproduce the issue with this system test.

def test_dbapi_fetchall(self):
    query = 'SELECT * FROM UNNEST([(1, 2), (3, 4), (5, 6)])'

    for arraysize in range(1, 5):
        Config.CURSOR.execute(query)
        self.assertEqual(Config.CURSOR.rowcount, 3, "expected 3 rows")
        Config.CURSOR.arraysize = arraysize
        rows = Config.CURSOR.fetchall()
        self.assertEqual(rows, [(1, 2), (3, 4), (5, 6)])

When arraysize is 1 (the default), only one result is returned when three are expected.

The unit tests are passing, but they mock out the core iterator class used, so I'm thinking the issue lies there or in how it is used.

I think the problem is that HTTPIterator is looking for nextPageToken but the getQueryResults response populates pageToken.

I'm able to reproduce this issue (only the first page is fetched) in the core API, not just the DB-API. By modifying the system test with 1000 rows and setting the page size to 100, I'm able to reproduce.

def test_large_query_w_public_data(self):
    PUBLIC = 'bigquery-public-data'
    DATASET_NAME = 'samples'
    TABLE_NAME = 'natality'
    LIMIT = 1000
    SQL = 'SELECT * from `{}.{}.{}` LIMIT {}'.format(
        PUBLIC, DATASET_NAME, TABLE_NAME, LIMIT)

    query = Config.CLIENT.run_sync_query(SQL)
    query.use_legacy_sql = False
    query.run()

    iterator = query.fetch_data(max_results=100)
    rows = list(iterator)
    self.assertEqual(len(rows), LIMIT)

Working to fix in this branch: https://github.com/tswast/google-cloud-python/tree/bq-dbapi/bigquery

Thank you for such a quick fix!

Fix merged. It should be in the next release. (I don't know when that will be)

Update dependency google-cloud-scheduler to v1.3.0 [(#3840)](GoogleCl…

Navigation Menu

Repository navigation

Fetching next batch of results using BigQuery's dbapi #3840

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Fetching next batch of results using BigQuery's dbapi #3840

Description

Activity

tswast commented on Aug 18, 2017

mxmzdlv commented on Aug 18, 2017

tswast commented on Aug 18, 2017

tswast commented on Aug 18, 2017

mxmzdlv commented on Aug 18, 2017

tswast commented on Aug 18, 2017

tswast commented on Aug 18, 2017

tswast commented on Aug 19, 2017

tswast commented on Aug 19, 2017

mxmzdlv commented on Aug 19, 2017

tswast commented on Aug 21, 2017

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions