Export data to Bigtable (reverse ETL)

This document describes how you can set up reverse ETL (RETL) from BigQuery to Bigtable. You can do this by using the EXPORT DATA statement to export data from a BigQuery table to a Bigtable table.

You can use a RETL workflow to Bigtable to combine BigQuery's analytics capabilities with Bigtable's low latency and high throughput. This workflow lets you serve data to application users without exhausting quotas and limits on BigQuery.

Characteristics of Bigtable tables

Bigtable tables are different from BigQuery tables in several ways:

• Both Bigtable and BigQuery tables are made of rows, but a Bigtable row is made of row key and column families that have an arbitrary number of columns belonging to the same column family.
• Column families for a given table are created at table creation time but can also be added or removed later. When a column family is created, columns that belong to it don't need to be specified.
• Bigtable columns don't need to be defined ahead of time and can be used to store data in their name (also known as a qualifier) within data size limits within tables.
• Bigtable columns can have any binary value within data size limits within tables.
• Bigtable columns always have a temporal dimension (also known as version). Any number of values might be stored in any row for the same column as long as the timestamp is not identical.
• A Bigtable timestamp is measured in microseconds since Unix epoch time—for example, 0 represents 1970-01-01T00:00:00 UTC. Timestamps must be a non-negative number of microseconds with millisecond granularity (only multiples of 1000us are accepted). The default Bigtable timestamp is 0.
• Data in Bigtable is read by row key, multiple row keys, range of row keys, or by using a filter. At least one row key or row keys range is required in all types of read requests except for a full table scan.

For information about preparing BigQuery results for export to Bigtable, see Prepare query results for export.

Before you begin

You must create a Bigtable instance and a Bigtable table to receive the exported data.

Grant Identity and Access Management (IAM) roles that give users the necessary permissions to perform each task in this document.

Required roles

To get the permissions that you need to export BigQuery data to Bigtable, ask your administrator to grant you the following IAM roles on your project:

• Export data from a BigQuery table: BigQuery Data Viewer (roles/bigquery.dataViewer)
• Run an export job: BigQuery User (roles/bigquery.user)
• Write data to a Bigtable table: Bigtable User (roles/bigtable.user)

For more information about granting roles, see Manage access to projects, folders, and organizations.

You might also be able to get the required permissions through custom roles or other predefined roles.

Limitations

• Encoding is limited to BINARY and TEXT only.
• The destination Bigtable app profile must be configured with single-cluster routing and a low request priority level.
• The Bigtable app profile must be configured to route data to a Bigtable cluster colocated with the BigQuery dataset. For more information, see location considerations.
• Exports to Bigtable are only supported for BigQuery Enterprise or Enterprise Plus editions. BigQuery Standard edition and on-demand compute are not supported.

Location considerations

• If your BigQuery dataset is in a multi-region, your Bigtable app profile must be configured to route data to a Bigtable cluster within that multi-region. For example, if your BigQuery dataset is in the US multi-region, the Bigtable cluster can be located in the us-west1 (Oregon) region, which is within the United States.
• If your BigQuery dataset is in a single region, your Bigtable app profile must be configured to route data to a Bigtable cluster in the same region. For example, if your BigQuery dataset is in the asia-northeast1 (Tokyo) region, your Bigtable cluster must also be in the asia-northeast1 (Tokyo) region.

For more information, see Bigtable locations.

Supported BigQuery types

The following types of data are supported when they're written to Bigtable:

BigQuery type	Bigtable value written
BYTES	Exported as is.
STRING	Converted to BYTES.
INTEGER	If bigtable_options.column_families.encoding is set to BINARY, then the value is written in an 8 byte, big-endian format (most significant byte first). If bigtable_options.column_families.encoding is set to TEXT, then the value is written as a human-readable string representing a number.
FLOAT	Writes value in the IEEE 754 8-byte output format.
BOOLEAN	If bigtable_options.column_families.encoding is set to BINARY, then the value is written as a 1 byte value (false = 0x00 or true = 0x01). If bigtable_options.column_families.encoding is set to TEXT, then the value is written as a text ("true" or "false").
JSON	An exported column of JSON type is interpreted as a group of columns belonging to a specific Bigtable column family. Members of the JSON object are interpreted as columns and their values are to be written to Bigtable. The name of the column to be written can be adjusted using the bigtable_options configuration. For example: JSON '{"FIELD1": "VALUE1", "FIELD2": "VALUE2"}' as MY_COLUMN_FAMILY Where values VALUE1 and VALUE2 are written to Bigtable as columns FIELD1 and FIELD2 to the column family MY_COLUMN_FAMILY. Note: A JSON document nested in another JSON or STRUCT type is written to Bigtable as a string. Exporting a STRUCT value nested in another struct is subject to limitations explained in the Configure exports with bigtable_options section.
STRUCT	An exported column of STRUCT type is interpreted as a group of columns belonging to a specific Bigtable column family. Members of the struct are interpreted as columns and their values to be written to Bigtable. The name of the column to be written can be adjusted using the bigtable_options configuration. For example: STRUCT<FIELD1 STRING, FIELD2 INTEGER> as MY_COLUMN_FAMILY Where values FIELD1 and FIELD2 are written to Bigtable as columns FIELD1 and FIELD2 to the column family MY_COLUMN_FAMILY.

These supported data types are similar to reading from external Bigtable tables for BigQuery.

NULL values in Bigtable

NULL values in Bigtable have the following constraints:

• Bigtable has no analog for NULL values. Exporting a NULL value for a given column family and column in Bigtable deletes the present values from a Bigtable row.

• If a Bigtable value with a given row key, column family, column qualifier, and timestamp doesn't exist prior to the export, the exported NULL values have no effect on the Bigtable row.

• When exporting a NULL value of the STRUCT or JSON type, all column values belonging to the corresponding column family of the affected row are deleted. You should cast the NULL value to the STRUCT or JSON type in order for the SQL engine to attach a correct type to it. The following query deletes all data from column family column_family1 with a set of given rowkeys:

  EXPORT DATA OPTIONS (...) AS
  SELECT
    rowkey,
  CAST(NULL as STRUCT<INT64>) AS column_family1 FROM T
  

• Rows with NULL row keys are skipped during the export. The number of skipped rows is returned in export statistics to the caller.

Configure exports with bigtable_options

You can use the bigtable_options configuration during an export to bridge the differences between BigQuery and Bigtable storage models. The configuration is expressed in the form of a JSON string, as seen in the following example:

EXPORT DATA OPTIONS(
   uri="https://bigtable.googleapis.com/projects/PROJECT_ID/instances/INSTANCE_ID/appProfiles/APP_PROFILE_ID/tables/TABLE",
   bigtable_options = """{
     "columnFamilies": [{
       "familyId": "COLUMN_FAMILY_NAME",
       "encoding": "ENCODING_VALUE",
       "columns": [
         {
           "qualifierString": "BIGTABLE_COLUMN_QUALIFIER",
           ["qualifierEncoded": "BASE_64_ENCODED_VALUE",]
           "fieldName": "BIGQUERY_RESULT_FIELD_NAME"
         }
       ]
    }]
   }"""
)

The following table describes the possible fields used in a bigtable_options configuration:

Field name	Description
columnFamilies	An array of column family descriptors.
columnFamilies.familyId	Identifier of Bigtable column family.
columnFamilies.encoding	Value can be set to BINARY or TEXT. For information about how types are encoded, see Supported BigQuery types.
columnFamilies.columns	An array of Bigtable column mappings.
columnFamilies.columns.qualifierString	Optional: A Bigtable column qualifier. Specify this value if the column qualifier has no non-UTF-8 codes. The fields qualifierString and qualifierEncoding are mutually exclusive. If neither qualifierString nor qualifierEncoded are specified, fieldName is used as a column qualifier.
columnFamilies.columns.qualifierEncoded	Optional: Base64-encoded column qualifier. Similar to qualifierString in case the column qualifier must have non-UTF-8 codes.
columnFamilies.columns.fieldName	Required: BigQuery result set field name. Can be an empty string in certain cases. For an example of how an empty fieldName value is used with fields of simple types, see Prepare query results for export.

Prepare query results for export

To export query results to Bigtable, the results must meet the following requirements:

• The result set must contain a column rowkey of the type STRING or BYTES.
• Row keys, column qualifiers, values, and timestamps must not exceed Bigtable data size limits within tables.
• At least one column other than rowkey must be present in the result set.
• Each result set column must be of one of the supported BigQuery types. Any unsupported column types must be converted to one of the supported types before exporting to Bigtable.

Bigtable doesn't require column qualifiers to be valid BigQuery column names, and Bigtable supports using any bytes. For information about overriding target column qualifiers for an export, see Configure exports with bigtable_options.

If you use exported values with Bigtable APIs, such as ReadModifyWriteRow, any numerical values must use the correct binary encoding.

By default, standalone result columns of types other than STRUCT or JSON are interpreted as values for destination column families equal to result column name, and column qualifier equal to an empty string.

To demonstrate how these data types are written, consider the follow SQL example, where column and column2 are standalone result columns:

SELECT
  x as column1, y as column2
FROM table

In this example query, SELECT x as column1 writes values to Bigtable under the column1 column family and '' (empty string) column qualifier when handling types other than JSON or STRUCT.

You can change how these types are written in an export using the bigtable_options configuration, as seen in the following example:

EXPORT DATA OPTIONS (
  …
  bigtable_options="""{
   "columnFamilies" : [
      {
        "familyId": "ordered_at",
        "columns": [
           {"qualifierString": "order_time", "fieldName": ""}
        ]
      }
   ]
}"""
) AS
SELECT
  order_id as rowkey,
  STRUCT(product, amount) AS sales_info,
  EXTRACT (MILLISECOND FROM order_timestamp AT TIME ZONE "UTC") AS ordered_at
FROM T

In this example, BigQuery table T contains the following row:

order_id	order_timestamp	product	amount
101	2023-03-28T10:40:54Z	Joystick	2

If you use the preceding bigtable_options configuration with table T, the following data is written to Bigtable:

rowkey	sales_info (column family)				ordered_at (column family)
101	product		amount		order_time
	1970-01-01T00:00:00Z	Joystick	1970-01-01T00:00:00Z	2	1680000054000

1680000054000 represents 2023-03-28T10:40:54Z in milliseconds since Unix epoch time in the UTC time zone.

Set timestamp for all cells in a row using _CHANGE_TIMESTAMP

You can add a _CHANGE_TIMESTAMP column of the TIMESTAMP type to the result for export. Every cell written to Bigtable uses the timestamp value from the _CHANGE_TIMESTAMP of the exported result row.

Bigtable doesn't support timestamps earlier than Unix epoch (1970-01-01T00:00:00Z). If _CHANGE_TIMESTAMP value is NULL, the Unix epoch time of 0 is used as the default timestamp value.

The following query writes cells for product and amount columns with the timestamp specified in the order_timestamp column of table T.

EXPORT DATA OPTIONS (...) AS
SELECT
  rowkey,
  STRUCT(product, amount) AS sales_info,
  order_timestamp as _CHANGE_TIMESTAMP
FROM T

Export continuously

If you want to continually process an export query, you can configure it as a continuous query.

Export multiple results with the same rowkey value

When you export a result containing multiple rows with the same rowkey value, values written to Bigtable end up in the same Bigtable row.

You can use this method to generate multiple versions of column values in the same row. In this example, the orders table in BigQuery contains the following data:

id	customer	order_timestamp	amount_spent
100	Bob	2023-01-01T10:10:54Z	10.99
101	Alice	2023-01-02T12:10:50Z	102.7
102	Bob	2023-01-04T15:17:01Z	11.1

The user then executes the following EXPORT DATA statement:

EXPORT DATA OPTIONS (
uri="https://bigtable.googleapis.com/projects/PROJECT-ID/instances/INSTANCE-ID/appProfiles/APP_PROFILE_ID/tables/TABLE",
format="CLOUD_BIGTABLE"
) AS
SELECT customer as rowkey, STRUCT(amount_spent) as orders_column_family, order_timestamp as _CHANGE_TIMESTAMP
FROM orders

Using this statement with the BigQuery orders table results in the following data written to Bigtable:

	orders_column_family
Row key	amount_spent
Alice	2023-01-02T12:10:50Z	102.7
Bob	2023-01-01T10:10:54Z	10.99
	2023-01-04T15:17:01Z	11.1

Exporting to Bigtable merges new values into the table instead of replacing entire rows. If values are already present in Bigtable for a row key, then new values can partially or fully override earlier values depending on the column family, column names, and timestamps of the cells being written.

Caution: Avoid exporting results that have multiple rows with different values for the same row key, column family, column qualifier, and timestamp. The outcome of such an export is non-deterministic and depends on the query plan and scheduling inside BigQuery. BigQuery cannot determine whether the value overridden during the export existed in Bigtable before export or was previously inserted by the same export process.

Export multiple columns as Protocol Buffer (Protobuf) values

Protocol buffers provide a flexible and efficient mechanism for serializing structured data. Exporting as a Protobuf can be beneficial considering how different types are handled between BigQuery and Bigtable. You can use BigQuery user-defined functions (UDFs) to export data as Protobuf binary values to Bigtable. For more information, see Export data as Protobuf columns.

Export optimization

You can change the throughput at which records are exported from BigQuery to Bigtable by modifying the number of nodes in the Bigtable destination cluster. The throughput (rows written per second) linearly scales with the numbers of nodes in the destination cluster. For example, if you double the number of nodes in your destination cluster, your export throughput will roughly double.

Pricing

When you export data in a standard query, you are billed using data extraction pricing. When you export data in a continuous query, you are billed using BigQuery capacity compute pricing. To run continuous queries, you must have a reservation that uses the Enterprise or Enterprise Plus edition, and a reservation assignment that uses the CONTINUOUS job type.

After the data is exported, you're charged for storing the data in Bigtable. For more information, see Bigtable pricing.