Pre-General Availability Draft: 2017-05-26
A common table expression (CTE) is a named temporary result set that exists within the scope of a single statement and that can be referred to later within that statement, possibly multiple times. The following discussion describes how to write statements that use CTEs.
For information about CTE optimization, see Section 126.96.36.199, “Optimizing Derived Tables, View References, and Common Table Expressions”.
To specify common table expressions, use a
WITH clause that has one or
more comma-separated subclauses, each of which provides a
subquery that produces a result set and a name associated with
the subquery. The following example defines CTEs named
cte2 in the
WITH clause, and refers to them
in the top-level
WITH cte1 AS (SELECT a, b FROM table1), cte2 AS (SELECT c, d FROM table2) SELECT b, d FROM cte1 JOIN cte2 WHERE cte1.a = cte2.c;
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: WITH [RECURSIVE]
col_name] ...)] AS (
col_name] ...)] AS (
cte_name names a single common
table expression and can be used as a table reference in the
statement containing the
subquery part of
( 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:
If a parenthesized list of names follows the CTE name, those names are the column names:
WITH cte (col1, col2) AS ( SELECT 1, 2 UNION ALL SELECT 3, 4 ) SELECT col1, col2 FROM cte;
The number of names in the list must be the same as the number of columns in the result set.
Otherwise, the column names come from the select list of the first
WITH cte AS ( SELECT 1 AS col1, 2 AS col2 UNION ALL SELECT 3, 4 ) SELECT col1, col2 FROM cte;
WITH clause is permitted in
WITH ... SELECT ... WITH ... UPDATE ... WITH ... DELETE ...
At the beginning of subqueries (including derived table subqueries):
SELECT ... WHERE id IN (WITH ... SELECT ...) ... SELECT * FROM (WITH ... SELECT ...) AS dt ...
INSERT ... WITH ... SELECT ... REPLACE ... WITH ... SELECT ... CREATE TABLE ... WITH ... SELECT ... CREATE VIEW ... WITH ... SELECT ... DECLARE CURSOR ... WITH ... SELECT ... EXPLAIN ... WITH ... SELECT ...
WITH cte1 AS (...) WITH cte2 AS (...) SELECT ...
However, a statement can contain multiple
WITH clauses if they occur at
WITH cte1 AS (SELECT 1) SELECT * FROM (WITH cte2 AS (SELECT 2) SELECT * FROM cte2 JOIN cte1) AS dt;
WITH clause can define one or
more common table expressions, but each CTE name must be
unique to the clause. This is illegal:
WITH cte1 AS (...), cte1 AS (...) SELECT ...
To make the statement legal, define the CTEs with unique names:
WITH cte1 AS (...), cte2 AS (...) SELECT ...
A CTE can refer to itself or to other CTEs:
A self-referencing CTE is recursive.
A CTE can refer to CTEs defined earlier in the same
WITHclause, but not those defined later.
This constraint rules out mutually-recursive CTEs, where
cte1because one of those references must be to a CTE defined later.
A CTE in a given query block can refer to CTEs defined in query blocks at a more outer level, but not CTEs defined in query blocks at a more inner level.
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.
A CTE cannot contain outer references. As with derived tables, which also prohibit outer references, this is a MySQL restriction, 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:
WITH RECURSIVE cte (n) AS ( SELECT 1 UNION ALL SELECT n + 1 FROM cte WHERE n < 5 ) SELECT * FROM cte;
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:
WITHclause must begin with
WITH RECURSIVEif any CTE in the
WITHclause refers to itself. (If no CTE refers to itself,
RECURSIVEis permitted but not required.)
If you forget
RECURSIVEfor a recursive CTE, this error is a likely result:
ERROR 1146 (42S02): Table '
cte_name' doesn't exist
SELECT ... -- return initial row set UNION ALL SELECT ... -- return additional row sets
SELECTproduces the initial row or rows for the CTE and does not refer to the CTE name. The second
SELECTproduces additional rows and recurses by referring to the CTE name in its
FROMclause. Recursion ends when this part produces no new rows. Thus, a recursive CTE consists of a nonrecursive
SELECTpart followed by a recursive
The types of the CTE result columns are inferred from the column types of the nonrecursive
SELECTpart only, and the columns are all nullable. For type determination, the recursive
SELECTpart is ignored.
If the nonrecursive and recursive parts are separated by
UNION DISTINCT, duplicate rows are eliminated. This is useful for queries that perform transitive closures, to avoid infinite loops.
Each iteration of the recursive part operates only on the rows produced by the previous iteration. If the recursive part has multiple query blocks, they all operate for each iteration on the rows produced by the previous iteration, and their results are concatenated for the next iteration.
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:
SELECT n + 1 FROM cte WHERE n < 5
At each iteration, that
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
the second iteration operates on the first iteration's row set
2) and produces
and so forth. This continues until recursion ends, 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:
WITH RECURSIVE cte AS ( SELECT 1 AS n, 'abc' AS str UNION ALL SELECT n + 1, CONCAT(str, str) FROM cte WHERE n < 3 ) SELECT * FROM cte;
In nonstrict SQL mode, the statement produces this output:
+------+------+ | n | str | +------+------+ | 1 | abc | | 2 | abc | | 3 | abc | +------+------+
In strict SQL mode, the statement produces an error:
ERROR 1406 (22001): Data too long for column 'str' at row 1
WITH RECURSIVE cte AS ( SELECT 1 AS n, CAST('abc' AS CHAR(20)) AS str UNION ALL SELECT n + 1, CONCAT(str, str) FROM cte WHERE n < 3 ) SELECT * FROM cte;
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:
WITH RECURSIVE cte AS ( SELECT 1 AS n, 1 AS p, -1 AS q UNION ALL SELECT n + 1, q * 2, p * 2 FROM cte WHERE n < 5 ) SELECT * FROM cte;
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:
SELECTpart must not contain these constructs:
Aggregate functions such as
SELECTpart must reference the CTE only once and only in its
FROMclause, not in any subquery. It can reference tables other than the CTE and join them with the CTE. If used in a join like this, the CTE must not be on the right side of a
These constraints come from the SQL standard, other than the
MySQL-specific exclusions of
It is important for recursive CTEs that the recursive
SELECT part include a condition
to terminate recursion. Otherwise, recursion terminates only
if some kind of resource exhaustion occurs. 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:
WITH RECURSIVE cte (n) AS ( SELECT 1 UNION ALL SELECT n + 1 FROM cte ) SELECT * FROM cte;
To guard against infinite recursion, set a per-session timeout by executing a statement like this prior to executing the CTE statement:
SET max_execution_time = 1000; -- impose one second timeout
Alternatively, include an optimizer hint within the CTE statement itself:
WITH RECURSIVE cte (n) AS ( SELECT 1 UNION ALL SELECT n + 1 FROM cte ) SELECT /*+ MAX_EXECUTION_TIME(1000) */ * FROM cte;
If a recursive query without an execution time limit enters an
infinite loop, you can terminate it from another session using
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.
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
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 9.4.4, “Internal Temporary Table Use in MySQL”.
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.
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:
WITH RECURSIVE fibonacci (n, fib_n, next_fib_n) AS ( SELECT 1, 0, 1 UNION ALL SELECT n + 1, next_fib_n, fib_n + next_fib_n FROM fibonacci WHERE n < 10 ) SELECT * FROM fibonacci;
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:
nis a display column to indicate that the row contains the
n-th Fibonacci number. For example, the 8th Fibonacci number is 13.
fib_ncolumn displays Fibonacci number
next_fib_ncolumn displays the next Fibonacci number after number
n. This column provides the next series value to the next row, so that row can produce the sum of the two previous series values in its
Recursion ends when
nreaches 10. This is an arbitrary choice, to limit the output to a small set of rows.
The preceding output shows the entire CTE result. To select
just part of it, add an appropriate
clause to the top-level
For example, to select the 8th Fibonacci number, do this:
WITH RECURSIVE fibonacci ......
SELECT fib_n FROM fibonacci WHERE n = 8;+-------+ | fib_n | +-------+ | 13 | +-------+
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:
SELECT * FROM sales ORDER BY date, price;+------------+--------+ | date | price | +------------+--------+ | 2017-01-03 | 100.00 | | 2017-01-03 | 200.00 | | 2017-01-06 | 50.00 | | 2017-01-08 | 10.00 | | 2017-01-08 | 20.00 | | 2017-01-08 | 150.00 | | 2017-01-10 | 5.00 | +------------+--------+
This query summarizes the sales per day:
SELECT date, SUM(price) AS sum_price
GROUP BY date
ORDER BY date;+------------+-----------+ | date | sum_price | +------------+-----------+ | 2017-01-03 | 300.00 | | 2017-01-06 | 50.00 | | 2017-01-08 | 180.00 | | 2017-01-10 | 5.00 | +------------+-----------+
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 produces
using a recursive CTE to generate that set of dates, joined
LEFT JOIN to the sales data.
Here is the CTE to generate the date range series:
WITH RECURSIVE dates (date) AS ( SELECT MIN(date) FROM sales UNION ALL SELECT date + INTERVAL 1 DAY FROM dates WHERE date + INTERVAL 1 DAY <= (SELECT MAX(date) FROM sales) ) SELECT * FROM dates;
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
sales table produces the sales summary
with a row for each date in the range:
WITH RECURSIVE dates (date) AS ( SELECT MIN(date) FROM sales UNION ALL SELECT date + INTERVAL 1 DAY FROM dates WHERE date + INTERVAL 1 DAY <= (SELECT MAX(date) FROM sales) ) SELECT dates.date, COALESCE(SUM(price), 0) AS sum_price FROM dates LEFT JOIN sales ON dates.date = sales.date GROUP BY dates.date ORDER BY dates.date;
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:
Are the queries inefficient, particularly the one with the
MAX()subquery executed for each row in the recursive
SELECT? Checking with
EXPLAINshows that the subqueries are optimized away for efficiency.
The use of
sum_pricecolumn on days for which no sales data occur in the
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).
CREATE TABLE employees ( id INT PRIMARY KEY NOT NULL, name VARCHAR(100) NOT NULL, manager_id INT NULL, INDEX (manager_id), FOREIGN KEY (manager_id) REFERENCES EMPLOYEES (id) ); INSERT INTO employees VALUES (333, "Yasmina", NULL), # Yasmina is the CEO (manager_id is NULL) (198, "John", 333), # John has ID 198 and reports to 333 (Yasmina) (692, "Tarek", 333), (29, "Pedro", 198), (4610, "Sarah", 29), (72, "Pierre", 29), (123, "Adil", 692);
The resulting data set looks like this:
SELECT * FROM employees ORDER BY id;+------+---------+------------+ | id | name | manager_id | +------+---------+------------+ | 29 | Pedro | 198 | | 72 | Pierre | 29 | | 123 | Adil | 692 | | 198 | John | 333 | | 333 | Yasmina | NULL | | 692 | Tarek | 333 | | 4610 | Sarah | 29 | +------+---------+------------+
To produce the organizational chart with the management chain for each employee (that is, the path from CEO to employee), use a recursive CTE:
WITH RECURSIVE employee_paths (id, name, path) AS ( SELECT id, name, CAST(id AS CHAR(200)) FROM employees WHERE manager_id IS NULL UNION ALL SELECT e.id, e.name, CONCAT(ep.path, ',', e.id) FROM employee_paths AS ep JOIN employees AS e ON ep.id = e.manager_id ) SELECT * FROM employee_paths ORDER BY path;
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:
SELECTproduces the row for the CEO (the row with a
pathcolumn is widened to
CHAR(200)to ensure that there is room for the longer
pathvalues produced by the recursive
Each row produced by the recursive
SELECTfinds all employees who report directly to an employee produced by a previous row. For each such employee, the row includes the employee ID and name, and the employee management chain. The chain is the manager's chain, with the employee ID added to the end.
Recursion ends when employees have no others who report to them.
WITH RECURSIVE ......
SELECT * FROM employees_extended
WHERE id IN (692, 4610)
ORDER BY path;+------+-------+-----------------+ | id | name | path | +------+-------+-----------------+ | 4610 | Sarah | 333,198,29,4610 | | 692 | Tarek | 333,692 | +------+-------+-----------------+
Common table expressions (CTEs) are similar to derived tables in some ways:
Both constructs are named.
Both constructs exist for the scope of a single statement.
Because of these similarities, CTEs and derived tables often can be used interchangeably. As a trivial example, these statements are equivalent:
WITH cte AS (SELECT 1) SELECT * FROM cte; SELECT * FROM (SELECT 1) AS dt;
However, CTEs have some advantages over derived tables:
A derived table can be referenced only a single time within a query. A CTE can be referenced multiple times. To use multiple instances of a derived table result, you must derive the result multiple times.
A CTE can be self-referencing (recursive).
One CTE can refer to another.
A CTE may be easier to read when its definition appears at the beginning of the statement rather than embedded within it.
CTEs are similar to tables created with
[TEMPORARY] TABLE but need not be defined or dropped
explicitly. For a CTE, you need no privileges to create