DECLARE keyword
DECLARE specifies a series of variable bindings used throughout your query.
A variable can hold a scalar value, a value list for an IN
filter such as symbol IN @symbols, or a bind variable. In a view definition,
a variable can be a parameter that the caller overrides or that an audited view
records.
This syntax is supported within SELECT queries.
Syntax
DECLARE @variable := expression [, @variable := expression ...]
[WITH ...]
SELECT ...
DECLARE [OVERRIDABLE] [AUDITED] @variable := expression
[, [OVERRIDABLE] [AUDITED] @variable := expression ...]
[WITH ...]
SELECT ...
DECLARE @variable := ( expression , [ expression [, expression ...] [,] ] )
WHERE expression [ NOT ] IN { @variable | $n | ( item [, item ...] ) }
A value list declares the set of values an IN filter tests
against. It has at least one comma. Its members are expressions, including
bind variables, but not lists or list variables. In IN, an item is a value
or a list variable. A $n bound to a text array must be the whole list: see
Lists of bind variables.
The OVERRIDABLE keyword only takes effect inside a
view definition. It
marks a variable as a parameter that the caller of the view can override at
query time. Variables without OVERRIDABLE use the value set in the view and
cannot be changed by the caller.
The AUDITED keyword only takes effect inside the definition of an
audited view, a QuestDB Enterprise feature. It
marks a variable whose resolved value each read of the view records in the
audit trail. It is independent of OVERRIDABLE, and the two may appear in
either order.
Mechanics
The DECLARE keyword comes before the SELECT clause in your query:
DECLARE
@x := 5
SELECT @x;
Use the variable binding operator := (walrus) to associate expressions to names.
It is easy to accidentally omit the : when writing variable binding expressions.
Don't confuse the := operator with a simple equality =!
You should see an error message like this:
expected variable assignment operator
:=
The above example declares a single binding, which states that the variable @x is replaced with the constant integer 5.
The variables are substituted at parse time: each reference is replaced by the variable's expression, so the variable is no longer present when the query is compiled.
So the above example reduces to this simple query:
SELECT 5;
| 5 |
|---|
| 5 |
Multiple bindings
To declare multiple variables, set the bind expressions with commas ,:
DECLARE
@x := 5,
@y := 2
SELECT @x + @y;
| column |
|---|
| 7 |
Variables as functions
A variable need not be just a constant. It could also be a function call, and variables with function values can be nested:
DECLARE
@today := today(),
@start := interval_start(@today),
@end := interval_end(@today)
SELECT @today = interval(@start, @end);
| column |
|---|
| true |
Declarations in subqueries
Declarations made in parent queries are available in subqueries.
DECLARE
@x := 5
SELECT y FROM (
SELECT @x AS y
);
| y |
|---|
| 5 |
Shadowing
If a subquery declares a variable of the same name, then the variable is shadowed and takes on the new value.
However, any queries above this subquery are unaffected - the variable bind is not globally mutated.
DECLARE
@x := 5
SELECT @x + y FROM (
DECLARE @x := 10
SELECT @x AS y
);
| column |
|---|
| 15 |
Declarations as subqueries
Declarations themselves can be subqueries.
We suggest that this is not overused, as removing the subquery definition from its execution location may make queries harder to debug.
Nevertheless, it is possible to define a variable as a subquery:
DECLARE
@subquery := (SELECT timestamp FROM trades)
SELECT * FROM @subquery;
You can even use already-declared variables to define your subquery variable:
DECLARE
@timestamp := timestamp,
@symbol := symbol,
@subquery := (SELECT @timestamp, @symbol FROM trades)
SELECT * FROM @subquery;
Declarations in CTEs
Naturally, DECLARE also works with CTEs:
DECLARE
@x := 5
WITH first AS (
DECLARE @x := 10
SELECT @x as a -- a = 10
),
second AS (
DECLARE @y := 4
SELECT
@x + @y as b, -- b = 5 + 4 = 9
a -- a = 10
FROM first
)
SELECT a, b
FROM second;
| a | b |
|---|---|
| 10 | 9 |
Bind variables
DECLARE syntax will work with prepared statements over PG Wire, so long as the client library
does not perform syntax validation that rejects the DECLARE syntax:
DECLARE @x := $1, @y := $2
SELECT @x::int + @y::int;
-- Then bind the following values: (1, 2)
| column |
|---|
| 3 |
Write bind variables as $1, $2, and so on. JDBC rewrites ? placeholders
to that form before it sends the query, but the server does not accept a raw
?.
DECLARE can also minimise repeated bind variables.
For example, rather than passing the same value to multiple positional arguments, you could instead use a declared variable and send a single bind variable:
-- instead of this:
SELECT $1 as name, id FROM users WHERE name = $2;
-- do this:
DECLARE @name := $1
SELECT @name as name, id FROM users WHERE name = @name;
A bind variable carries a value, not a column name. To repeat a column, declare the column itself, and bind only the values:
DECLARE
@col := price,
@symbol := $1
SELECT avg(@col), min(@col), max(@col)
FROM trades
WHERE symbol = @symbol;
Value lists
A parenthesised, comma-separated list declares the values of an IN filter
once, so that a query or a view can name the set instead of spelling it out:
DECLARE @symbols := ('BTC-USDT', 'ETH-USDT')
SELECT timestamp, symbol, price, amount
FROM trades
WHERE symbol IN @symbols AND timestamp IN '$now-1h..$now';
DECLARE @symbols := ('BTC-USDT', 'ETH-USDT')
SELECT timestamp, symbol, price, amount
FROM trades
WHERE symbol NOT IN @symbols AND timestamp IN '$now-1h..$now';
The list is expanded into the IN when the query is parsed, so the first query
above is exactly symbol IN ('BTC-USDT', 'ETH-USDT'). Each member keeps its
own type, and every form of IN works as it does with a written-out list,
including NOT IN and interval scans on the designated timestamp.
IN @symbolsandIN (@symbols)are equivalent.- A list mixes with literals:
symbol IN ('SOL-USDT', @symbols). - One list variable can be assigned to another:
@majors := @symbols. - A list of one needs a trailing comma,
('BTC-USDT',). Without it,('BTC-USDT')is a parenthesised value. - A bracketed subquery,
(SELECT ...)or(DECLARE ... SELECT ...), is a subquery and not a list.
Lists of timestamps
A list of timestamps follows the rules of a written-out IN, which depend on
the list's length:
- With two or more members, each member is one instant.
('2026-01-05', '2026-01-12')matches only rows at midnight on those days, not the whole days. - A list of one is an interval, like a single string.
('2026-01-05',)matches the whole day.
So a view parameter that a caller overrides with one date instead of two changes from instants to a whole day. To match whole days, declare a TICK date list, which is a single string:
DECLARE @days := '[2026-01-05,2026-01-12,2026-01-19]'
SELECT timestamp, symbol, price, amount
FROM trades
WHERE timestamp IN @days;
Lists of bind variables
To filter a SYMBOL, VARCHAR or STRING value on any number of values with
one prepared statement, bind one text array over the PostgreSQL wire protocol:
SELECT timestamp, symbol, price FROM trades WHERE symbol IN $1;
-- bind $1 as a text array, such as {BTC-USDT,ETH-USDT,SOL-USDT}
The array can also be bound through a variable:
DECLARE @symbols := $1
SELECT timestamp, symbol, price FROM trades WHERE symbol IN @symbols;
With JDBC, write ? in place of $1, and bind
ps.setArray(1, connection.createArrayOf("varchar", symbols)). With psql,
\bind '{BTC-USDT,ETH-USDT,SOL-USDT}'.
The array must be the whole IN list: IN $1, IN ($1) or NOT IN $1. It
cannot be mixed with other values, as in IN ('ETH-USDT', $1): put those
values in the array instead. A typed array bind, as from JDBC, fails there with
STRING constant expected. An untyped one, as from psql \bind, is taken as
one string, such as '{SOL-USDT,BTC-USDT}', so the query silently returns
wrong results.
A view parameter can take an array bind too, but only when
the view's body uses it as the whole list, as symbol IN @symbols does.
Otherwise, bind a list of bind variables, as below.
Members of a list can also be bind variables. An AUDITED parameter of an
audited view needs this form, because it
cannot hold an array:
DECLARE @symbols := ($1, $2)
SELECT timestamp, symbol, price FROM trades WHERE symbol IN @symbols;
The statement text fixes the number of members: ($1, $2) always takes two
values, so prepare one statement for each count. A single bind variable needs
the trailing comma, ($1,). With a client that sends parameters untyped, such
as psql \bind, it also needs a cast, ($1::varchar,): untyped, ($1,) is
taken as one array bind variable for the whole list. An audited view refuses
both an array and an untyped ($1,) with
audited view parameter has a type that cannot be audited.
Lists in views
In a view, an OVERRIDABLE list
can be overridden with a list of a different length:
CREATE VIEW trades_for AS (
DECLARE OVERRIDABLE @symbols := ('BTC-USDT', 'ETH-USDT')
SELECT timestamp, symbol, price FROM trades WHERE symbol IN @symbols
);
DECLARE @symbols := ('SOL-USDT',) SELECT * FROM trades_for;
Errors
A list has no value of its own, so it can only be used on the right-hand side
of IN. These fail:
SELECT @symbols,WHERE symbol = @symbols,OVER (PARTITION BY @symbols)or@all := (@symbols, 'SOL-USDT'):
declared list can only be used on the right-hand side of IN@x := ('BTC-USDT', ('ETH-USDT', 'SOL-USDT')):
nested lists are not supported, list members have to be values@x := ():
value expected in list
To combine a list with more values, write both in the IN:
symbol IN (@symbols, 'SOL-USDT').
Limitations
Most basic expressions are supported, and we provide examples later in this document.
We suggest you use variables to simplify repeated constants within your code, and minimise how many places you need to update the constant.
Disallowed expressions
However, not all expressions are supported. The following are explicitly disallowed:
SQL statement fragments
DECLARE
@x := FROM trades
SELECT 5 @x;
-- table and column names that are SQL keywords have to be enclosed in double quotes, such as "FROM"```
Language client support
Some language SQL clients do not allow identifiers to be passed as if it was a normal value. One example is psycopg.
In this case, you should use an alternate API to splice in identifiers, for example:
cur.execute(
sql.SQL("""
DECLARE @col := {}
SELECT max(@col), min(@col), avg(price)
FROM btc_trades;
""").format(sql.Identifier('price')))
Examples
SAMPLE BY
DECLARE
@period := 1m,
@window := '2024-11-25',
@symbol := 'ETH-USDT'
SELECT
timestamp, symbol, side, sum(amount) as volume
FROM trades
WHERE side = 'sell'
AND timestamp IN @window
AND symbol = @symbol
SAMPLE BY @period
FILL(NULL);
| timestamp | symbol | side | volume |
|---|---|---|---|
| 2024-11-25T00:00:00.000000Z | ETH-USDT | sell | 153.470574999999 |
| 2024-11-25T00:01:00.000000Z | ETH-USDT | sell | 298.927738 |
| 2024-11-25T00:02:00.000000Z | ETH-USDT | sell | 66.253058 |
| ... | ... | ... | ... |
INSERT INTO SELECT
INSERT INTO trades (timestamp, symbol)
SELECT * FROM
(
DECLARE
@x := now(),
@y := 'ETH-USDT'
SELECT @x as timestamp, @y as symbol
);
CREATE TABLE AS SELECT
CREATE TABLE trades AS (
DECLARE
@x := now(),
@y := 'ETH-USDT'
SELECT @x as timestamp, @y as symbol, 123 as price
);