Structured row key queries
If a table has a row key schema that defines structured row keys, you can use SQL to query the individual segments – or columns – of the row key.
Continuous materialized views, including asynchronous secondary indexes, generate structured row keys by default. You can also define a structured row key for any Bigtable table that you want to query with SQL by creating a row key schema for the table. For more information, see Manage row key schemas.
Keys
In a Bigtable table without a structured row key schema, each row is indexed by a single row key. When you query the table with SQL, the row key is a column named _key
that SQL uses as the primary key. It's not a pseudocolumn, so values in the _key
column are returned when you execute a SELECT *
query on the table.
On the other hand, in a table that has a row key schema, the row key (primary key) is the combination of all the columns named as fields in the row key schema. When you add a row key schema to a table, the _key
column becomes a pseudocolumn, which means it won't show up in a SELECT *
query, but you can select it explicitly with a SELECT _key
statement.
Sample queries
The examples in this section assume that a table named sales
has the following schema:
field { field_name: "user_id" type: { bytes_type { encoding { raw {} } } } } field { field_name: "purchase_date" type: { string_type { encoding { utf8_bytes {} } } } } field { field_name: "order_number" type: { string_type { encoding { utf8_bytes {} } } } } encoding { delimited_bytes { delimiter "#" } }
The sales
table contains the following data. The product
column family has two columns.
row key | product | |
---|---|---|
product_type | product_name | |
"user1#2025-05-20#abcd1233" | "phone" | "iPhone16_pro_max" |
"user1#2025-05-20#abcd1235" | "shoes" | "nike_hyperdunk" |
"user2#2025-05-24#defg456" | "headphones" | "sony_wh_1000mx5" |
Structured row key query results
Because the sales
table has structured row keys, if you query the table with a SELECT *
statement, the query returns each segment of the row key as a separate column. As in any SQL query to a Bigtable table, columns in a column family are expressed as maps.
SELECT * from sales
Results look like the following:
user_id | purchase_date | order_number | product |
---|---|---|---|
user1 | 2025-05-20 | "abcd1233" | { product_type: "phone", product_name: "iPhone16_pro_max" } |
user1 | 2025-05-20 | "abcd1235" | { product_type: "shoes", product_name: "nike_hyperdunk" } |
user2 | 2025-05-24 | "defg456" | { product_type: "headphones", product_name: "sony_wh_1000mx5" } |
You can also specify the row key columns in your query, as shown in the following example:
SELECT product[product_type] AS product_type, product[product_name] AS product_name FROM sales WHERE user_id = b"user1"
The results look like the following:
product_type | product_name |
---|---|
"phone" | "iphone16_pro_max" |
"shoes" | "nike_hyperdunk" |
Filters
You can filter on the row key schema columns using SQL functions. The following example assumes that CURRENT_DATE()
returns 2025-05-24
:
SELECT user_id, product["product_name"] AS product_name FROM sales WHERE PARSE_DATE("YYYY-MM-DD", purchase_date) = CURRENT_DATE() AND user_id = b"user2"
The results are as follows:
user_id | product_name |
---|---|
user2 | "sony_wh_1000mx5" |
Aggregate queries
The following example shows how use an aggregate query on structured row key fields:
SELECT user_id, product[product_type] AS product_type, count(*) AS count FROM sales GROUP BY 1, 2
The query results are as follows:
user_id | product_type | count |
---|---|---|
user1 | phone | 1 |
user1 | shoes | 1 |
user2 | headphones | 1 |
Original row key
To retrieve the original row key in a table with structured row keys, specify the _key
column in your query.
SELECT _key, user_id FROM sales
The query returns the following:
_key | user_id |
---|---|
"user1#2025-05-20#abcd1233" | user1 |
"user1#2025-05-20#abcd1235" | user1 |
"user2#2025-05-24#defg456" | user2 |