13.2.13 WITH Syntax (Common Table Expressions)

To specify common table expressions, use a WITH clause that has one or more comma-separated subclauses. Each subclause provides a subquery that produces a result set, and associates a name with the subquery. The following example defines CTEs named cte1 and cte2 in the WITH clause, and refers to them in the top-level SELECT that follows the WITH clause:

In the statement containing the WITH clause, each CTE name can be referenced to access the corresponding CTE result set.

A CTE name can be referenced in other CTEs, enabling CTEs to be defined based on other CTEs.

A CTE can refer to itself to define a recursive CTE. Common applications of recursive CTEs include series generation and traversal of hierarchical or tree-structured data.

Common table expressions are an optional part of the syntax for DML statements. They are defined using a WITH clause:

cte_name names a single common table expression and can be used as a table reference in the statement containing the WITH clause.

The subquery part of AS (subquery) is called the “subquery of the CTE” and is what produces the CTE result set. The parentheses following AS are required.

A common table expression is recursive if its subquery refers to its own name. The RECURSIVE keyword must be included if any CTE in the WITH clause is recursive. For more information, see Recursive Common Table Expressions.

Determination of column names for a given CTE occurs as follows:

A WITH clause is permitted in these contexts:

Only one WITH clause is permitted at the same level. WITH followed by WITH at the same level is not permitted, so this is illegal:

To make the statement legal, use a single WITH clause that separates the subclauses by a comma:

However, a statement can contain multiple WITH clauses if they occur at different levels:

A WITH clause can define one or more common table expressions, but each CTE name must be unique to the clause. This is illegal:

To make the statement legal, define the CTEs with unique names:

A CTE can refer to itself or to other CTEs:

For resolving references to objects with the same names, derived tables hide CTEs; and CTEs hide base tables, TEMPORARY tables, and views. Name resolution occurs by searching for objects in the same query block, then proceeding to outer blocks in turn while no object with the name is found.

Like derived tables, a CTE cannot contain outer references prior to MySQL 8.0.14. This is a MySQL restriction that is lifted in MySQL 8.0.14, not a restriction of the SQL standard. For additional syntax considerations specific to recursive CTEs, see Recursive Common Table Expressions.

A recursive common table expression is one having a subquery that refers to its own name. For example:

When executed, the statement produces this result, a single column containing a simple linear sequence:

+------+
| n    |
+------+
|    1 |
|    2 |
|    3 |
|    4 |
|    5 |
+------+

A recursive CTE has this structure:

The recursive CTE subquery shown earlier has this nonrecursive part that retrieves a single row to produce the initial row set:

The CTE subquery also has this recursive part:

At each iteration, that SELECT produces a row with a new value one greater than the value of n from the previous row set. The first iteration operates on the initial row set (1) and produces 1+1=2; the second iteration operates on the first iteration's row set (2) and produces 2+1=3; and so forth. This continues until recursion ends, which occurs when n is no longer less than 5.

If the recursive part of a CTE produces wider values for a column than the nonrecursive part, it may be necessary to widen the column in the nonrecursive part to avoid data truncation. Consider this statement:

In nonstrict SQL mode, the statement produces this output:

+------+------+
| n    | str  |
+------+------+
|    1 | abc  |
|    2 | abc  |
|    3 | abc  |
+------+------+

The str column values are all 'abc' because the nonrecursive SELECT determines the column widths. Consequently, the wider str values produced by the recursive SELECT are truncated.

In strict SQL mode, the statement produces an error:

ERROR 1406 (22001): Data too long for column 'str' at row 1

To address this issue, so that the statement does not produce truncation or errors, use CAST() in the nonrecursive SELECT to make the str column wider:

Now the statement produces this result, without truncation:

+------+--------------+
| n    | str          |
+------+--------------+
|    1 | abc          |
|    2 | abcabc       |
|    3 | abcabcabcabc |
+------+--------------+

Columns are accessed by name, not position, which means that columns in the recursive part can access columns in the nonrecursive part that have a different position, as this CTE illustrates:

Because p in one row is derived from q in the previous row, and vice versa, the positive and negative values values swap positions in each successive row of the output:

+------+------+------+
| n    | p    | q    |
+------+------+------+
|    1 |    1 |   -1 |
|    2 |   -2 |    2 |
|    3 |    4 |   -4 |
|    4 |   -8 |    8 |
|    5 |   16 |  -16 |
+------+------+------+

Some syntax constraints apply within recursive CTE subqueries:

These constraints come from the SQL standard, other than the MySQL-specific exclusions of ORDER BY, LIMIT, and DISTINCT.

For recursive CTEs, EXPLAIN output rows for recursive SELECT parts display Recursive in the Extra column.

Cost estimates displayed by EXPLAIN represent cost per iteration, which might differ considerably from total cost. The optimizer cannot predict the number of iterations because it cannot predict when the WHERE clause will become false.

CTE actual cost may also be affected by result set size. A CTE that produces many rows may require an internal temporary table large enough to be converted from in-memory to on-disk format and may suffer a performance penalty. If so, increasing the permitted in-memory temporary table size may improve performance; see Section 8.4.4, “Internal Temporary Table Use in MySQL”.

It is important for recursive CTEs that the recursive SELECT part include a condition to terminate recursion. As a development technique to guard against a runaway recursive CTE, you can force termination by placing a limit on execution time:

Suppose that a recursive CTE is mistakenly written with no recursion execution termination condition:

By default, cte_max_recursion_depth has a value of 1000, causing the CTE to terminate when it recurses past 1000 levels. Applications can change the session value to adjust for their requirements:

You can also set the global cte_max_recursion_depth value to affect all sessions that begin subsequently.

For queries that execute and thus recurse slowly or in contexts for which there is reason to set the cte_max_recursion_depth value very high, another way to guard against deep recursion is to set a per-session timeout. To do so, execute a statement like this prior to executing the CTE statement:

Alternatively, include an optimizer hint within the CTE statement itself:

If a recursive query without an execution time limit enters an infinite loop, you can terminate it from another session using KILL QUERY. Within the session itself, the client program used to run the query might provide a way to kill the query. For example, in mysql, typing Control+C interrupts the current statement.

As mentioned previously, recursive common table expressions (CTEs) are frequently used for series generation and traversing hierarchical or tree-structured data. This section shows some simple examples of these techniques.

Table des matières Haut

Fibonacci Series Generation

A Fibonacci series begins with the two numbers 0 and 1 (or 1 and 1) and each number after that is the sum of the previous two numbers. A recursive common table expression can generate a Fibonacci series if each row produced by the recursive SELECT has access to the two previous numbers from the series. The following CTE generates a 10-number series using 0 and 1 as the first two numbers:

The CTE produces this result:

+------+-------+------------+
| n    | fib_n | next_fib_n |
+------+-------+------------+
|    1 |     0 |          1 |
|    2 |     1 |          1 |
|    3 |     1 |          2 |
|    4 |     2 |          3 |
|    5 |     3 |          5 |
|    6 |     5 |          8 |
|    7 |     8 |         13 |
|    8 |    13 |         21 |
|    9 |    21 |         34 |
|   10 |    34 |         55 |
+------+-------+------------+

How the CTE works:

The preceding output shows the entire CTE result. To select just part of it, add an appropriate WHERE clause to the top-level SELECT. For example, to select the 8th Fibonacci number, do this:

Table des matières Haut

Date Series Generation

A common table expression can generate a series of successive dates, which is useful for generating summaries that include a row for all dates in the series, including dates not represented in the summarized data.

Suppose that a table of sales numbers contains these rows:

This query summarizes the sales per day:

However, that result contains “holes” for dates not represented in the range of dates spanned by the table. A result that represents all dates in the range can be produced using a recursive CTE to generate that set of dates, joined with a LEFT JOIN to the sales data.

Here is the CTE to generate the date range series:

The CTE produces this result:

+------------+
| date       |
+------------+
| 2017-01-03 |
| 2017-01-04 |
| 2017-01-05 |
| 2017-01-06 |
| 2017-01-07 |
| 2017-01-08 |
| 2017-01-09 |
| 2017-01-10 |
+------------+

How the CTE works:

Joining the CTE with a LEFT JOIN against the sales table produces the sales summary with a row for each date in the range:

The output looks like this:

+------------+-----------+
| date       | sum_price |
+------------+-----------+
| 2017-01-03 |    300.00 |
| 2017-01-04 |      0.00 |
| 2017-01-05 |      0.00 |
| 2017-01-06 |     50.00 |
| 2017-01-07 |      0.00 |
| 2017-01-08 |    180.00 |
| 2017-01-09 |      0.00 |
| 2017-01-10 |      5.00 |
+------------+-----------+

Some points to note:

Table des matières Haut

Hierarchical Data Traversal

Recursive common table expressions are useful for traversing data that forms a hierarchy. Consider these statements that create a small data set that shows, for each employee in a company, the employee name and ID number, and the ID of the employee's manager. The top-level employee (the CEO), has a manager ID of NULL (no manager).

The resulting data set looks like this:

To produce the organizational chart with the management chain for each employee (that is, the path from CEO to employee), use a recursive CTE:

The CTE produces this output:

+------+---------+-----------------+
| id   | name    | path            |
+------+---------+-----------------+
|  333 | Yasmina | 333             |
|  198 | John    | 333,198         |
|   29 | Pedro   | 333,198,29      |
| 4610 | Sarah   | 333,198,29,4610 |
|   72 | Pierre  | 333,198,29,72   |
|  692 | Tarek   | 333,692         |
|  123 | Adil    | 333,692,123     |
+------+---------+-----------------+

How the CTE works:

To find the path for a specific employee or employees, add a WHERE clause to the top-level SELECT. For example, to display the results for Tarek and Sarah, modify that SELECT like this:

Common table expressions (CTEs) are similar to derived tables in some ways:

Because of these similarities, CTEs and derived tables often can be used interchangeably. As a trivial example, these statements are equivalent:

However, CTEs have some advantages over derived tables:

CTEs are similar to tables created with CREATE [TEMPORARY] TABLE but need not be defined or dropped explicitly. For a CTE, you need no privileges to create tables.