Cockroach Labs will stop providing Assistance Support for this version on May 12, 2021. Prior to that date, upgrade to a more recent version to continue receiving support. For more details, see the Release Support Policy.

SQL subqueries enable reuse of the results from a selection query within another query.


CockroachDB supports two kinds of subqueries:

Data writes in subqueries

When a subquery contains a data-modifying statement (INSERT, DELETE, etc.), the data modification is always executed to completion even if the surrounding query only uses a subset of the result rows.

This is true both for subqueries defined using the (...) or [...] notations, and those defined using WITH.

For example:

   FROM [INSERT INTO t(x) VALUES (1), (2), (3) RETURNING x]
  LIMIT 1;

This query always inserts 3 rows into t, even though the surrounding query only observes 1 row using LIMIT.

Correlated subqueries

CockroachDB's cost-based optimizer supports most correlated subqueries.

A subquery is said to be "correlated" when it uses table or column names defined in the surrounding query.

For example, to find every customer with at least one order, run:

      customers AS c
          SELECT * FROM orders AS o WHERE o.customer_id = c.id

The subquery is correlated because it uses c defined in the surrounding query.


The cost-based optimizer supports most correlated subqueries, with the exception of correlated subqueries that generate side effects inside a CASE statement.

Performance best practices


CockroachDB is currently undergoing major changes to evolve and improve the performance of subqueries. The restrictions and workarounds listed in this section will be lifted or made unnecessary over time.

  • Scalar subqueries currently disable the distribution of the execution of a query. To ensure maximum performance on queries that process a large number of rows, make the client application compute the subquery results ahead of time and pass these results directly in the surrounding query.

  • The results of scalar subqueries are currently loaded entirely into memory when the execution of the surrounding query starts. To prevent execution errors due to memory exhaustion, ensure that subqueries return as few results as possible.

See also

YesYes NoNo