Load data with cross-cloud operations

As a BigQuery administrator or analyst, you can load data from an Amazon Simple Storage Service (Amazon S3) bucket or Azure Blob Storage into BigQuery tables. You can either join the transferred data with the data present in Google Cloud regions or take advantage of BigQuery features like BigQuery ML.

You can transfer data into BigQuery in the following ways:

Quotas and limits

For information about quotas and limits, see query jobs quotas and limits.

Pricing

You are billed for the bytes that are transferred across clouds by using the LOAD statement. For pricing information, see Data transfer pricing.

You are billed for the bytes that are transferred across clouds by using the CREATE TABLE AS SELECT statement or INSERT INTO SELECT statement and for the compute capacity.

Both LOAD and CREATE TABLE AS SELECT statements require slots in the BigQuery Omni regions to scan Amazon S3 and Blob Storage files to load them.

Before you begin

To provide Google Cloud with read access to the files in other clouds, ask your administrator to create a connection and share it with you. For information about how to create connections, see Connect to Amazon S3 or Blob Storage.

Required role

To get the permissions that you need to load data using cross-cloud transfers, ask your administrator to grant you the BigQuery Data Editor (roles/bigquery.dataEditor) IAM role on the dataset. For more information about granting roles, see Manage access.

This predefined role contains the permissions required to load data using cross-cloud transfers. To see the exact permissions that are required, expand the Required permissions section:

Required permissions

The following permissions are required to load data using cross-cloud transfers:

  • bigquery.tables.create
  • bigquery.tables.get
  • bigquery.tables.updateData
  • bigquery.tables.update
  • bigquery.jobs.create
  • bigquery.connections.use

You might also be able to get these permissions with custom roles or other predefined roles.

For more information about IAM roles in BigQuery, see Predefined roles and permissions.

Load data

You can load data into BigQuery with the LOAD DATA [INTO|OVERWRITE] statement.

Limitations

  • Loading data into a destination table with ingestion time partition is not supported.
  • LOAD DATA jobs don't run on reservations. Jobs utilize on-demand slots that are managed by Google Cloud.
  • The connection and the destination dataset must belong to the same project. Loading data across projects is not supported.
  • LOAD DATA is only supported when you transfer data from an Amazon Simple Storage Service (Amazon S3) or Azure Blob Storage to a colocated BigQuery region. For more information, see Locations.
    • You can transfer data from any US region to a US multi-region. You can also transfer from any EU region to a EU multi-region.

Example

Example 1

The following example loads a parquet file named sample.parquet from an Amazon S3 bucket into the test_parquet table with an auto-detect schema:

LOAD DATA INTO mydataset.testparquet
  FROM FILES (
    uris = ['s3://test-bucket/sample.parquet'],
    format = 'PARQUET'
  )
  WITH CONNECTION `aws-us-east-1.test-connection`

Example 2

The following example loads a CSV file with the prefix sampled* from your Blob Storage into the test_csv table with predefined column partitioning by time:

LOAD DATA INTO mydataset.test_csv (Number INT64, Name STRING, Time DATE)
  PARTITION BY Time
  FROM FILES (
    format = 'CSV', uris = ['azure://test.blob.core.windows.net/container/sampled*'],
    skip_leading_rows=1
  )
  WITH CONNECTION `azure-eastus2.test-connection`

Example 3

The following example overwrites the existing table test_parquet with data from a file named sample.parquet with an auto-detect schema:

LOAD DATA OVERWRITE mydataset.testparquet
  FROM FILES (
    uris = ['s3://test-bucket/sample.parquet'],
    format = 'PARQUET'
  )
  WITH CONNECTION `aws-us-east-1.test-connection`

Filter data

You can filter data before transferring them into BigQuery by using the CREATE TABLE AS SELECT statement and the INSERT INTO SELECT statement.

Limitations

  • If the result of the SELECT query exceeds 20 GiB in logical bytes, the query fails. The table is not created and data is not transferred. To learn how to reduce the size of data that is scanned, see Reduce data processed in queries.

  • Temporary tables are not supported.

  • Transferring the Well-known binary (WKB) geospatial data format is not supported.

  • INSERT INTO SELECT statement does not support transferring data into clustered table.

  • In the INSERT INTO SELECT statement, if the destination table is the same as the source table in the SELECT query, then the INSERT INTO SELECT statement doesn't modify any rows in the destination table. The destination table isn't modified as BigQuery can't read data across regions.

  • CREATE TABLE AS SELECT and INSERT INTO SELECT are only supported when you transfer data from an Amazon S3 or Blob Storage to a colocated BigQuery region. For more information, see Locations.

    • You can transfer data from any US region to a US multi-region. You can also transfer from any EU region to a EU multi-region.

Example

Example 1

Suppose you have a BigLake table named myawsdataset.orders that references data from Amazon S3. You want to transfer data from that table to a BigQuery table myotherdataset.shipments in the US multi-region.

First, display information about the myawsdataset.orders table:

    bq show myawsdataset.orders;

The output is similar to the following:

  Last modified             Schema              Type     Total URIs   Expiration
----------------- -------------------------- ---------- ------------ -----------
  31 Oct 17:40:28   |- l_orderkey: integer     EXTERNAL   1
                    |- l_partkey: integer
                    |- l_suppkey: integer
                    |- l_linenumber: integer
                    |- l_returnflag: string
                    |- l_linestatus: string
                    |- l_commitdate: date

Next, display information about the myotherdataset.shipments table:

  bq show myotherdataset.shipments

The output is similar to the following. Some columns are omitted to simplify the output.

  Last modified             Schema             Total Rows   Total Bytes   Expiration   Time Partitioning   Clustered Fields   Total Logical
 ----------------- --------------------------- ------------ ------------- ------------ ------------------- ------------------ ---------------
  31 Oct 17:34:31   |- l_orderkey: integer      3086653      210767042                                                         210767042
                    |- l_partkey: integer
                    |- l_suppkey: integer
                    |- l_commitdate: date
                    |- l_shipdate: date
                    |- l_receiptdate: date
                    |- l_shipinstruct: string
                    |- l_shipmode: string

Now, using the CREATE TABLE AS SELECT statement you can selectively load data to the myotherdataset.orders table in the US multi-region:

CREATE OR REPLACE TABLE
  myotherdataset.orders
  PARTITION BY DATE_TRUNC(l_commitdate, YEAR) AS
SELECT
  *
FROM
  myawsdataset.orders
WHERE
  EXTRACT(YEAR FROM l_commitdate) = 1992;

You can then perform a join operation with the newly created table:

SELECT
  orders.l_orderkey,
  orders.l_orderkey,
  orders.l_suppkey,
  orders.l_commitdate,
  orders.l_returnflag,
  shipments.l_shipmode,
  shipments.l_shipinstruct
FROM
  myotherdataset.shipments
JOIN
  `myotherdataset.orders` as orders
ON
  orders.l_orderkey = shipments.l_orderkey
AND orders.l_partkey = shipments.l_partkey
AND orders.l_suppkey = shipments.l_suppkey
WHERE orders.l_returnflag = 'R'; -- 'R' means refunded.

When new data is available, append the data of the 1993 year to the destination table using the INSERT INTO SELECT statement:

INSERT INTO
   myotherdataset.orders
 SELECT
   *
 FROM
   myawsdataset.orders
 WHERE
   EXTRACT(YEAR FROM l_commitdate) = 1993;

Example 2

The following example inserts data into an ingestion-time partitioned table:

CREATE TABLE
 mydataset.orders(id String, numeric_id INT)
PARTITION BY _PARTITIONDATE;

After creating a partitioned table, you can insert data into the ingestion-time partitioned table:

INSERT INTO
 mydataset.orders(
   _PARTITIONTIME,
   id,
   numeric_id)
SELECT
 TIMESTAMP("2023-01-01"),
 id,
 numeric_id,
FROM
 mydataset.ordersof23
WHERE
 numeric_id > 4000000;

Best practices

  • Avoid loading multiple files that are less than 5 MB. Instead, create an external table for your file and export query result to Amazon S3 or Blob Storage to create a larger file. This method helps to improve the transfer time of your data.
  • For information about the limit for maximum query result, see BigQuery Omni maximum query result size.
  • If your source data is in a gzip-compressed file, then while creating external tables, set the external_table_options.compression option to GZIP.

What's next