After learning about SQL routines from the introduction, the following sections show numerous examples of valid SQL routines. The routines are suitable as inline routines or catalog routines, after adjusting the name and adjusting the example invocations.

The examples combine numerous supported statements. Refer to the specific statement documentation for further details:

A very simple routine that returns a static value without requiring any input:

FUNCTION answer()
RETURNS BIGINT
RETURN 42

​
Inline and catalog routines

A full example of this routine as inline routine and usage in a string concatenation with a cast:

WITH
  FUNCTION answer()
  RETURNS BIGINT
  RETURN 42
SELECT 'The answer is ' || CAST(answer() as varchar);
-- The answer is 42

Provided the catalog example supports routine storage in the default schema, you can use the following:

USE example.default;
CREATE FUNCTION example.default.answer()
  RETURNS BIGINT
  RETURN 42;

With the routine stored in the catalog, you can run the routine multiple times without repeated definition:

SELECT example.default.answer() + 1; -- 43
SELECT 'The answer is' || CAST(example.default.answer() as varchar); -- The answer is 42

​
Declaration examples

The result of calling the routine answer() is always identical, so you can declare it as deterministic, and add some other information:

FUNCTION answer()
LANGUAGE SQL
DETERMINISTIC
RETURNS BIGINT
COMMENT 'Provide the answer to the question about life, the universe, and everything.'
RETURN 42

A simple routine that returns a greeting back to the input string fullname concatenating two strings and the input value:

FUNCTION hello(fullname VARCHAR)
RETURNS VARCHAR
RETURN 'Hello, ' || fullname || '!'

Following is an example invocation:

SELECT hello('Jane Doe'); -- Hello, Jane Doe!

A first example routine, that uses multiple statements in a BEGIN block. It calculates the result of a multiplication of the input integer with 99. The bigint data type is used for all variables and values. The value of integer 99 is cast to bigint in the default value assignment for the variable x.

FUNCTION times_ninety_nine(a bigint)
RETURNS bigint
BEGIN
  DECLARE x bigint DEFAULT CAST(99 AS bigint);
  RETURN x * a;
END

Following is an example invocation:

SELECT times_ninety_nine(CAST(2 as bigint)); -- 198

​
Conditional flows

A first example of conditional flow control in a routine using the CASE statement. The simple bigint input value is compared to a number of values.

FUNCTION simple_case(a bigint)
RETURNS varchar
BEGIN
  CASE a
    WHEN 0 THEN RETURN 'zero';
    WHEN 1 THEN RETURN 'one';
    WHEN 10 THEN RETURN 'ten';
    WHEN 20 THEN RETURN 'twenty';
    ELSE RETURN 'other';
  END CASE;
  RETURN NULL;
END

Following are a couple of example invocations with result and explanation:

SELECT simple_case(0); -- zero
SELECT simple_case(1); -- one
SELECT simple_case(-1); -- other (from else clause)
SELECT simple_case(10); -- ten
SELECT simple_case(11); -- other (from else clause)
SELECT simple_case(20); -- twenty
SELECT simple_case(100); -- other (from else clause)
SELECT simple_case(null); -- null .. but really??

A second example of a routine with a CASE statement, this time with two parameters, showcasing the importance of the order of the conditions.

FUNCTION search_case(a bigint, b bigint)
RETURNS varchar
BEGIN
  CASE
    WHEN a = 0 THEN RETURN 'zero';
    WHEN b = 1 THEN RETURN 'one';
    WHEN a = DECIMAL '10.0' THEN RETURN 'ten';
    WHEN b = 20.0E0 THEN RETURN 'twenty';
    ELSE RETURN 'other';
  END CASE;
  RETURN NULL;
END

Following are a couple of example invocations with result and explanation:

SELECT search_case(0,0); -- zero
SELECT search_case(1,1); -- one
SELECT search_case(0,1); -- zero (not one since the second check is never reached)
SELECT search_case(10,1); -- one (not ten since the third check is never reached)
SELECT search_case(10,2); -- ten
SELECT search_case(10,20); -- ten (not twenty)
SELECT search_case(0,20); -- zero (not twenty)
SELECT search_case(3,20); -- twenty
SELECT search_case(3,21); -- other
SELECT simple_case(null,null); -- null .. but really??

​
Fibonacci example

This routine calculates the n-th value in the Fibonacci series, in which each number is the sum of the two preceding ones. The two initial values are set to 1 as the defaults for a and b. The routine uses an IF statement condition to return 1 for all input values of 2 or less. The WHILE block then starts to calculate each number in the series, starting with a=1 and b=1 and iterates until it reaches the n-th position. In each iteration is sets a and b for the preceding to values, so it can calculate the sum, and finally return it. Note that processing the routine takes longer and longer with higher n values, and the result is deterministic.

FUNCTION fib(n bigint)
RETURNS bigint
BEGIN
  DECLARE a, b bigint DEFAULT 1;
  DECLARE c bigint;
  IF n <= 2 THEN
    RETURN 1;
  END IF;
  WHILE n > 2 DO
    SET n = n - 1;
    SET c = a + b;
    SET a = b;
    SET b = c;
  END WHILE;
  RETURN c;
END

Following are a couple of example invocations with result and explanation:

SELECT fib(-1); -- 1
SELECT fib(0); -- 1
SELECT fib(1); -- 1
SELECT fib(2); -- 1
SELECT fib(3); -- 2
SELECT fib(4); -- 3
SELECT fib(5); -- 5
SELECT fib(6); -- 8
SELECT fib(7); -- 13
SELECT fib(8); -- 21

​
Labels and loops

This routine uses the top label to name the WHILE block, and then controls the flow with conditional statements, ITERATE, and LEAVE. For the values of a=1 and a=2 in the first two iterations of the loop the ITERATE call moves the flow up to top before b is ever increased. Then b is increased for the values a=3, a=4, a=5, a=6, and a=7, resulting in b=5. The LEAVE call then causes the exit of the block before a is increased further to 10 and therefore the result of the routine is 5.

FUNCTION labels()
RETURNS bigint
BEGIN
  DECLARE a, b int DEFAULT 0;
  top: WHILE a < 10 DO
    SET a = a + 1;
    IF a < 3 THEN
      ITERATE top;
    END IF;
    SET b = b + 1;
    IF a > 6 THEN
      LEAVE top;
    END IF;
  END WHILE;
  RETURN b;
END

This routine implements calculating the n to the power of p by repeated multiplication and keeping track of the number of multiplications performed. Note that this routine does not return the correct 0 for p=0 since the top block is merely escaped and the value of n is returned. The same incorrect behavior happens for negative values of p:

FUNCTION power(n int, p int)
RETURNS int
  BEGIN
    DECLARE r int DEFAULT n;
    top: LOOP
      IF p <= 1 THEN
        LEAVE top;
      END IF;
      SET r = r * n;
      SET p = p - 1;
    END LOOP;
    RETURN r;
  END

Following are a couple of example invocations with result and explanation:

SELECT power(2, 2); -- 4
SELECT power(2, 8); -- 256
SELECT power(3, 3); -- 256
SELECT power(3, 0); -- 3, which is wrong
SELECT power(3, -2); -- 3, which is wrong

This routine returns 7 as a result of the increase of b in the loop from a=3 to a=10:

FUNCTION test_repeat_continue()
RETURNS bigint
BEGIN
  DECLARE a int DEFAULT 0;
  DECLARE b int DEFAULT 0;
  top: REPEAT
    SET a = a + 1;
    IF a <= 3 THEN
      ITERATE top;
    END IF;
    SET b = b + 1;
  UNTIL a >= 10
  END REPEAT;
  RETURN b;
END

This routine returns 2 and shows that labels can be repeated and label usage within a block refers to the label of that block:

FUNCTION test()
RETURNS int
BEGIN
  DECLARE r int DEFAULT 0;
  abc: LOOP
    SET r = r + 1;
    LEAVE abc;
  END LOOP;
  abc: LOOP
    SET r = r + 1;
    LEAVE abc;
  END LOOP;
  RETURN r;
END

​
Routines and built-in functions

This routine show that multiple data types and built-in functions like length() and cardinality() can be used in a routine. The two nested BEGIN blocks also show how variable names are local within these blocks x, but the global r from the top-level block can be accessed in the nested blocks:

FUNCTION test()
RETURNS bigint
BEGIN
  DECLARE r bigint DEFAULT 0;
  BEGIN
    DECLARE x varchar DEFAULT 'hello';
    SET r = r + length(x);
  END;
  BEGIN
    DECLARE x array(int) DEFAULT array[1, 2, 3];
    SET r = r + cardinality(x);
  END;
  RETURN r;
END

​
Optional parameter example

Routines can invoke other routines and other functions. The full signature of a routine is composed of routine name and parameters, and determines the exact routine to use. You can declare multiple routines with the same name, but with different number of arguments or different argument types. One example use case is to implement an optional parameter.

The following routine truncates a string to the specified length including three dots at the end of the output:

FUNCTION dots(input varchar, length integer)
RETURNS varchar
BEGIN
  IF length(input) > length THEN
    RETURN substring(input, 1, length-3) || '...';
  END IF;
  RETURN input;
END;

Following are example invocations and output:

SELECT dots('A long string that will be shortened',15);
-- A long strin...
SELECT	dots('A short string',15);
-- A short string

If you want to provide a routine with the same name, but without the parameter for length, you can create another routine that invokes the preceding routine:

FUNCTION dots(input varchar)
RETURNS varchar
RETURN dots(input, 15);

You can now use both routines. When the length parameter is omitted the default value from the second declaration is used.

SELECT dots('A long string that will be shortened',15);
-- A long strin...
SELECT dots('A long string that will be shortened');
-- A long strin...
SELECT dots('A long string that will be shortened',20);
-- A long string tha...

​
Date string parsing example

This example routine parses a date string of type VARCHAR into TIMESTAMP WITH TIME ZONE. Date strings are commonly represented by ISO 8601 standard, such as 2023-12-01, 2023-12-01T23. Date strings are also often represented in the YYYYmmdd and YYYYmmddHH format, such as 20230101 and 2023010123. Hive tables can use this format to represent day and hourly partitions, for example /day=20230101, /hour=2023010123.

This routine parses date strings in a best-effort fashion and can be used as a replacement for date string manipulation functions such as date, date_parse, from_iso8601_date, and from_iso8601_timestamp.

Note that the routine defaults the time value to 00:00:00.000 and the time zone to the session time zone.

FUNCTION from_date_string(date_string VARCHAR)
RETURNS TIMESTAMP WITH TIME ZONE
BEGIN
  IF date_string like '%-%' THEN -- ISO 8601
    RETURN from_iso8601_timestamp(date_string);
  ELSEIF length(date_string) = 8 THEN -- YYYYmmdd
      RETURN date_parse(date_string, '%Y%m%d');
  ELSEIF length(date_string) = 10 THEN -- YYYYmmddHH
      RETURN date_parse(date_string, '%Y%m%d%H');
  END IF;
  RETURN NULL;
END

Following are a couple of example invocations with result and explanation:

SELECT from_date_string('2023-01-01'); -- 2023-01-01 00:00:00.000 UTC (using the ISO 8601 format)
SELECT from_date_string('2023-01-01T23'); -- 2023-01-01 23:00:00.000 UTC (using the ISO 8601 format)
SELECT from_date_string('2023-01-01T23:23:23'); -- 2023-01-01 23:23:23.000 UTC (using the ISO 8601 format)
SELECT from_date_string('20230101'); -- 2023-01-01 00:00:00.000 UTC (using the YYYYmmdd format)
SELECT from_date_string('2023010123'); -- 2023-01-01 23:00:00.000 UTC (using the YYYYmmddHH format)
SELECT from_date_string(NULL); -- NULL (handles NULL string)
SELECT from_date_string('abc'); -- NULL (not matched to any format)

​
Human readable days

Trino includes a built-in function called human_readable_seconds that formats a number of seconds into a string:

SELECT human_readable_seconds(134823);
-- 1 day, 13 hours, 27 minutes, 3 seconds

The example routine hrd formats a number of days into a human readable text that provides the approximate number of years and months:

FUNCTION hrd(d integer)
RETURNS VARCHAR
BEGIN
    DECLARE answer varchar default 'About ';
    DECLARE years real;
    DECLARE months real;
    SET years = truncate(d/365);
    IF years > 0 then
        SET answer = answer || format('%1.0f', years) || ' year';
    END IF;
    IF years > 1 THEN
        SET answer = answer || 's';
    END IF;
    SET d = d - cast( years AS integer) * 365 ;
    SET months = truncate(d / 30);
    IF months > 0 and years > 0 THEN
        SET answer = answer || ' and ';
    END IF;
    IF months > 0 THEN
        set answer = answer || format('%1.0f', months) || ' month';
    END IF;
    IF months > 1 THEN
        SET answer = answer || 's';
    END IF;
    IF years < 1 and months < 1 THEN
        SET answer = 'Less than 1 month';
    END IF;
    RETURN answer;
END;

The following examples show the output for a range of values under one month, under one year, and various larger values:

SELECT hrd(10); -- Less than 1 month
SELECT hrd(95); -- About 3 months
SELECT hrd(400); -- About 1 year and 1 month
SELECT hrd(369); -- About 1 year
SELECT hrd(800); -- About 2 years and 2 months
SELECT hrd(1100); -- About 3 years
SELECT hrd(5000); -- About 13 years and 8 months

Improvements of the routine could include the following modifications:

  • Take into account that one month equals 30.4375 days.
  • Take into account that one year equals 365.25 days.
  • Add weeks to the output.
  • Expand the cover decades, centuries, and millenia.

​
Truncating long strings

This example routine strtrunc truncates strings longer than 60 characters, leaving the first 30 and the last 25 characters, and cutting out extra characters in the middle.

FUNCTION strtrunc(input VARCHAR)
RETURNS VARCHAR
RETURN
    CASE WHEN length(input) > 60
    THEN substr(input, 1, 30) || ' ... ' || substr(input, length(input) - 25)
    ELSE input
    END;

The preceding declaration is very compact and consists of only one complex statement with a CASE expression and multiple function calls. It can therefore define the complete logic in the RETURN clause.

The following statement shows the same capability within the routine itself. Note the duplicate RETURN inside and outside the CASE statement and the required END CASE;. The second RETURN statement is required, because a routine must end with a RETURN statement. As a result the ELSE clause can be omitted.

FUNCTION strtrunc(input VARCHAR)
RETURNS VARCHAR
BEGIN
    CASE WHEN length(input) > 60
    THEN
        RETURN substr(input, 1, 30) || ' ... ' || substr(input, length(input) - 25);
    ELSE
        RETURN input;
    END CASE;
    RETURN input;
END;

The next example changes over from a CASE to an IF statement, and avoids the duplicate RETURN:

FUNCTION strtrunc(input VARCHAR)
RETURNS VARCHAR
BEGIN
    IF length(input) > 60 THEN
        RETURN substr(input, 1, 30) || ' ... ' || substr(input, length(input) - 25);
    END IF;
    RETURN input;
END;

All the preceding examples create the same output. Following is an example query which generates long strings to truncate:

WITH
data AS (
    SELECT substring('strtrunc truncates strings longer than 60 characters, leaving the prefix and suffix visible', 1, s.num) AS value
    FROM table(sequence(start=>40, stop=>80, step=>5)) AS s(num)
)
SELECT
    data.value
  , strtrunc(data.value) AS truncated
FROM data
ORDER BY data.value;

The preceding query produces the following output with all variants of the routine:

                                      value                                       |                           truncated
----------------------------------------------------------------------------------+---------------------------------------------------------------
 strtrunc truncates strings longer than 6                                         | strtrunc truncates strings longer than 6
 strtrunc truncates strings longer than 60 cha                                    | strtrunc truncates strings longer than 60 cha
 strtrunc truncates strings longer than 60 characte                               | strtrunc truncates strings longer than 60 characte
 strtrunc truncates strings longer than 60 characters, l                          | strtrunc truncates strings longer than 60 characters, l
 strtrunc truncates strings longer than 60 characters, leavin                     | strtrunc truncates strings longer than 60 characters, leavin
 strtrunc truncates strings longer than 60 characters, leaving the                | strtrunc truncates strings lon ... 60 characters, leaving the
 strtrunc truncates strings longer than 60 characters, leaving the pref           | strtrunc truncates strings lon ... aracters, leaving the pref
 strtrunc truncates strings longer than 60 characters, leaving the prefix an      | strtrunc truncates strings lon ... ers, leaving the prefix an
 strtrunc truncates strings longer than 60 characters, leaving the prefix and suf | strtrunc truncates strings lon ... leaving the prefix and suf

A possible improvement is to introduce parameters for the total length.

​
Formatting bytes

Trino includes a built-in format_number() function. However it is using units that don’t work well with bytes. The following format_data_size routine can format large values of bytes into a human readable string.

FUNCTION format_data_size(input BIGINT)
RETURNS VARCHAR
  BEGIN
    DECLARE value DOUBLE DEFAULT CAST(input AS DOUBLE);
    DECLARE result BIGINT;
    DECLARE base INT DEFAULT 1024;
    DECLARE unit VARCHAR DEFAULT 'B';
    DECLARE format VARCHAR;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'kB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'MB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'GB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'TB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'PB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'EB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'ZB';
    END IF;
    IF abs(value) >= base THEN
      SET value = value / base;
      SET unit = 'YB';
    END IF;
    IF abs(value) < 10 THEN
      SET format = '%.2f';
    ELSEIF abs(value) < 100 THEN
      SET format = '%.1f';
    ELSE
      SET format = '%.0f';
    END IF;
    RETURN format(format, value) || unit;
  END;

Below is a query to show how it formats a wide range of values.

WITH
data AS (
    SELECT CAST(pow(10, s.p) AS BIGINT) AS num
    FROM table(sequence(start=>1, stop=>18)) AS s(p)
    UNION ALL
    SELECT -CAST(pow(10, s.p) AS BIGINT) AS num
    FROM table(sequence(start=>1, stop=>18)) AS s(p)
)
SELECT
    data.num
  , format_data_size(data.num) AS formatted
FROM data
ORDER BY data.num;

The preceding query produces the following output:

         num          | formatted
----------------------+-----------
 -1000000000000000000 | -888PB
  -100000000000000000 | -88.8PB
   -10000000000000000 | -8.88PB
    -1000000000000000 | -909TB
     -100000000000000 | -90.9TB
      -10000000000000 | -9.09TB
       -1000000000000 | -931GB
        -100000000000 | -93.1GB
         -10000000000 | -9.31GB
          -1000000000 | -954MB
           -100000000 | -95.4MB
            -10000000 | -9.54MB
             -1000000 | -977kB
              -100000 | -97.7kB
               -10000 | -9.77kB
                -1000 | -1000B
                 -100 | -100B
                  -10 | -10.0B
                    0 | 0.00B
                   10 | 10.0B
                  100 | 100B
                 1000 | 1000B
                10000 | 9.77kB
               100000 | 97.7kB
              1000000 | 977kB
             10000000 | 9.54MB
            100000000 | 95.4MB
           1000000000 | 954MB
          10000000000 | 9.31GB
         100000000000 | 93.1GB
        1000000000000 | 931GB
       10000000000000 | 9.09TB
      100000000000000 | 90.9TB
     1000000000000000 | 909TB
    10000000000000000 | 8.88PB
   100000000000000000 | 88.8PB
  1000000000000000000 | 888PB

​
Charts

Trino already has a built-in bar() color function, but it’s using ANSI escape codes to output colors, and thus is only usable for displaying results in a terminal. The following example shows a similar routine, that only uses ASCII characters.

FUNCTION ascii_bar(value DOUBLE)
RETURNS VARCHAR
BEGIN
  DECLARE max_width DOUBLE DEFAULT 40.0;
  RETURN array_join(
    repeat('█',
        greatest(0, CAST(floor(max_width * value) AS integer) - 1)), '')
        || ARRAY[' ', '▏', '▎', '▍', '▌', '▋', '▊', '▉', '█'][cast((value % (cast(1 as double) / max_width)) * max_width * 8 + 1 as int)];
END;

It can be used to visualize a value.

WITH
data AS (
    SELECT
        cast(s.num as double) / 100.0 AS x,
        sin(cast(s.num as double) / 100.0) AS y
    FROM table(sequence(start=>0, stop=>314, step=>10)) AS s(num)
)
SELECT
    data.x,
    round(data.y, 4) AS y,
    ascii_bar(data.y) AS chart
FROM data
ORDER BY data.x;

The preceding query produces the following output:

  x  |   y    |                  chart
-----+--------+-----------------------------------------
 0.0 |    0.0 |
 0.1 | 0.0998 | ███
 0.2 | 0.1987 | ███████
 0.3 | 0.2955 | ██████████▉
 0.4 | 0.3894 | ██████████████▋
 0.5 | 0.4794 | ██████████████████▏
 0.6 | 0.5646 | █████████████████████▋
 0.7 | 0.6442 | ████████████████████████▊
 0.8 | 0.7174 | ███████████████████████████▊
 0.9 | 0.7833 | ██████████████████████████████▍
 1.0 | 0.8415 | ████████████████████████████████▋
 1.1 | 0.8912 | ██████████████████████████████████▋
 1.2 |  0.932 | ████████████████████████████████████▎
 1.3 | 0.9636 | █████████████████████████████████████▌
 1.4 | 0.9854 | ██████████████████████████████████████▍
 1.5 | 0.9975 | ██████████████████████████████████████▉
 1.6 | 0.9996 | ███████████████████████████████████████
 1.7 | 0.9917 | ██████████████████████████████████████▋
 1.8 | 0.9738 | ██████████████████████████████████████
 1.9 | 0.9463 | ████████████████████████████████████▉
 2.0 | 0.9093 | ███████████████████████████████████▍
 2.1 | 0.8632 | █████████████████████████████████▌
 2.2 | 0.8085 | ███████████████████████████████▍
 2.3 | 0.7457 | ████████████████████████████▉
 2.4 | 0.6755 | ██████████████████████████
 2.5 | 0.5985 | ███████████████████████
 2.6 | 0.5155 | ███████████████████▋
 2.7 | 0.4274 | ████████████████▏
 2.8 |  0.335 | ████████████▍
 2.9 | 0.2392 | ████████▋
 3.0 | 0.1411 | ████▋
 3.1 | 0.0416 | ▋

It’s also possible to draw more compacted charts. Following is a routine drawing vertical bars:

FUNCTION vertical_bar(value DOUBLE)
RETURNS VARCHAR
RETURN ARRAY[' ', '▁', '▂', '▃', '▄', '▅', '▆', '▇', '█'][cast(value * 8 + 1 as int)];

It can be used to draw a distribution of values, in a single column.

WITH
measurements(sensor_id, recorded_at, value) AS (
    VALUES
        ('A', date '2023-01-01', 5.0)
      , ('A', date '2023-01-03', 7.0)
      , ('A', date '2023-01-04', 15.0)
      , ('A', date '2023-01-05', 14.0)
      , ('A', date '2023-01-08', 10.0)
      , ('A', date '2023-01-09', 1.0)
      , ('A', date '2023-01-10', 7.0)
      , ('A', date '2023-01-11', 8.0)
      , ('B', date '2023-01-03', 2.0)
      , ('B', date '2023-01-04', 3.0)
      , ('B', date '2023-01-05', 2.5)
      , ('B', date '2023-01-07', 2.75)
      , ('B', date '2023-01-09', 4.0)
      , ('B', date '2023-01-10', 1.5)
      , ('B', date '2023-01-11', 1.0)
),
days AS (
    SELECT date_add('day', s.num, date '2023-01-01') AS day
    -- table function arguments need to be constant but range could be calculated
    -- using: SELECT date_diff('day', max(recorded_at), min(recorded_at)) FROM measurements
    FROM table(sequence(start=>0, stop=>10)) AS s(num)
),
sensors(id) AS (VALUES ('A'), ('B')),
normalized AS (
    SELECT
        sensors.id AS sensor_id,
        days.day,
        value,
        value / max(value) OVER (PARTITION BY sensor_id) AS normalized
    FROM days
    CROSS JOIN sensors
    LEFT JOIN measurements m ON day = recorded_at AND m.sensor_id = sensors.id
)
SELECT
    sensor_id,
    min(day) AS start,
    max(day) AS stop,
    count(value) AS num_values,
    min(value) AS min_value,
    max(value) AS max_value,
    avg(value) AS avg_value,
    array_join(array_agg(coalesce(vertical_bar(normalized), ' ') ORDER BY day), '') AS distribution
FROM normalized
WHERE sensor_id IS NOT NULL
GROUP BY sensor_id
ORDER BY sensor_id;

The preceding query produces the following output:

 sensor_id |   start    |    stop    | num_values | min_value | max_value | avg_value | distribution
-----------+------------+------------+------------+-----------+-----------+-----------+--------------
 A         | 2023-01-01 | 2023-01-11 |          8 |      1.00 |     15.00 |      8.38 | ▃ ▄█▇  ▅▁▄▄
 B         | 2023-01-01 | 2023-01-11 |          7 |      1.00 |      4.00 |      2.39 |   ▄▆▅ ▆ █▃▂

​
Top-N

Trino already has a built-in aggregate function called approx_most_frequent(), that can calculate most frequently occurring values. It returns a map with values as keys and number of occurrences as values. Maps are not ordered, so when displayed, the entries can change places on subsequent runs of the same query, and readers must still compare all frequencies to find the one most frequent value. The following is a routine returns ordered results as a string.

FUNCTION format_topn(input map<varchar, bigint>)
RETURNS VARCHAR
NOT DETERMINISTIC
BEGIN
  DECLARE freq_separator VARCHAR DEFAULT '=';
  DECLARE entry_separator VARCHAR DEFAULT ', ';
  RETURN array_join(transform(
    reverse(array_sort(transform(
      transform(
        map_entries(input),
          r -> cast(r AS row(key varchar, value bigint))
      ),
      r -> cast(row(r.value, r.key) AS row(value bigint, key varchar)))
    )),
    r -> r.key || freq_separator || cast(r.value as varchar)),
    entry_separator);
END;

Following is an example query to count generated strings:

WITH
data AS (
    SELECT lpad('', 3, chr(65+(s.num / 3))) AS value
    FROM table(sequence(start=>1, stop=>10)) AS s(num)
),
aggregated AS (
    SELECT
        array_agg(data.value ORDER BY data.value) AS all_values,
        approx_most_frequent(3, data.value, 1000) AS top3
    FROM data
)
SELECT
    a.all_values,
    a.top3,
    format_topn(a.top3) AS top3_formatted
FROM aggregated a;

The preceding query produces the following result:

                     all_values                     |         top3          |    top3_formatted
----------------------------------------------------+-----------------------+---------------------
 [AAA, AAA, BBB, BBB, BBB, CCC, CCC, CCC, DDD, DDD] | {AAA=2, CCC=3, BBB=3} | CCC=3, BBB=3, AAA=2