avg(), count(), max(), min(), sum()

Each of the aggregate functions avg(), count(), max(), min(), and sum() is invoked by using the same syntax—either the GROUP BY syntax or the OVER syntax.

avg()

Signature:

input value:       smallint, int, bigint, numeric, real, double precision, interval

return value:      numeric, double precision, interval

Notes: The lists of input and return data types give the distinct kinds. Briefly, the input can be any of the data types for which the notion of average is meaningful—including, therefore, the data types whose values are constrained to be whole numbers. Because, for example, the average of the whole numbers 3 and 4 is 3.5, the return data type is never one whose values are constrained to be whole numbers. Here are the specific mappings:

INPUT             OUTPUT
----------------  ----------------
smallint          numeric
int               numeric
bigint            numeric
numeric           numeric
real              double precision
double precision  double precision
interval          interval

Purpose: Computes the arithmetic mean of a set of summable values by adding them all together and dividing by the number of values. If the set contains nulls, then these are simply ignored—both when computing the sum and when counting the number of values.

count()

Signature:

input value:       "any"

return value:      bigint

Purpose: Counts the number of non null values in a set. The data type of the values is of no consequence.

max(), min()

Signature: max() and min() have the same signature as each other.

input value:       smallint, int, bigint, numeric, real, double precision, money,
                   character, varchar, text,
                   date, abstime, time with time zone, time without time zone,
                   timestamp with time zone, timestamp without time zone, tid,
                   interval, oid, inet, anyenum, anyarray

return value:      < Same as the input value's data type except when the input is "varchar".
                     When the input is "varchar", the return is "text" >

Purpose: Computes the greatest, or the least, value among the values in the set using the rule that is used for the particular data type in the ORDER BY clause. nulls are removed before sorting the values.

sum()

Signature:

input value:       smallint, int, bigint, numeric, real, double precision, money, interval

return value:      bigint, numeric, double precision, real, money, interval

Notes: The lists of input and return data types give the distinct kinds. Briefly, the input can be any of the data types for which the notion of summation is meaningful. Here are the specific mappings:

INPUT             OUTPUT
----------------  ----------------
smallint          bigint
int               bigint
bigint            numeric
numeric           numeric
real              real
double precision  double precision
money             money
interval          interval

Purpose: Computes the sum of a set of summable values by adding them all together. If the set contains nulls, then these are simply ignored.

Examples

First create and populate the test table:

drop table if exists t cascade;
create table t(
  k     int      primary key,
  class int      not null,
  n     numeric,
  t     text     not null);

insert into t(k, class, n, t)
select
  (1 + s.v),
  case (s.v) < 5
    when true then 1
              else 2
  end,
  case (s.v) <> 4
    when true then (100.0 + s.v)::numeric
              else null
  end,
  chr(97 + s.v)
from generate_series(0, 9) as s(v);

\pset null <null>
select k, class, n, t from t order by k;

This is the result:

 k  | class |   n    | t
----+-------+--------+---
  1 |     1 |    100 | a
  2 |     1 |    101 | b
  3 |     1 |    102 | c
  4 |     1 |    103 | d
  5 |     1 | <null> | e
  6 |     2 |    105 | f
  7 |     2 |    106 | g
  8 |     2 |    107 | h
  9 |     2 |    108 | i
 10 |     2 |    109 | j

GROUP BY syntax

Now try this query:

select
  class,

  -- "numeric" arguments.
  to_char(avg(n), '990.99')      as "avg(n)",
  count(n)                       as "count(n)",
  min(n)                         as "min(n)",
  max(n)                         as "max(n)",
  sum(n)                         as "sum(n)",

  -- "text" arguments.
  count((chr(n::int)||t)::text)  as "count(expr)",
  min(t)                         as "min(t)",
  max(t)                         as "max(t)"
from t
group by class
order by class;

It produces this result:

 class | avg(n)  | count(n) | min(n) | max(n) | sum(n) | count(expr) | min(t) | max(t)
-------+---------+----------+--------+--------+--------+-------------+--------+--------
     1 |  101.50 |        4 |    100 |    103 |    406 |           4 | a      | e
     2 |  107.00 |        5 |    105 |    109 |    535 |           5 | f      | j

The results for "class = 1" show a generic property of aggregate functions: if, for some row, the expression that is used to invoke the function evaluates to null, then that row is silently ignored and the result is calculated by using only the not null values in the designated window.

Here's a more dramatic illustration of the same rule:

drop table if exists t cascade;
create table t2(
  k     int      primary key,
  n     numeric);

insert into t2(k, n)
select
  s.v,
  case
    when (s.v) between 1 and 3 then 1
    when (s.v) between 4 and 6 then 2
                               else null
  end
from generate_series(1, 9) as s(v);

\pset null <null>
select k, n from t2 order by k;

This is the result:

 k |   n
---+--------
 1 |      1
 2 |      1
 3 |      1
 4 |      2
 5 |      2
 6 |      2
 7 | <null>
 8 | <null>
 9 | <null>

Now try this:

select count(distinct n) from t2;

Here, the count() expression uses the distinct keyword. This is the result:

 count
-------
     2

Because null means, quite literally, the complete absence of information, it is excluded from the set of to-be-processed values.

OVER syntax

Here is a "mechanical" re-write of the GROUP BY syntax code example above.

select
  class,

  -- "numeric" arguments.
  n,
  to_char(avg(n) filter (where n not in (102, 107)) over w, '990.99')  as "avg(n)",
  count(n)       filter (where n not in (102, 107)) over w             as "count(n)",
  min(n)         filter (where n not in (102, 107)) over w             as "min(n)",
  max(n)         filter (where n not in (102, 107)) over w             as "max(n)",
  sum(n)         filter (where n not in (102, 107)) over w             as "sum(n)",

  -- "text" arguments.
  t,
  count((chr(n::int)||t)::text) over w                                 as "count(expr)",
  min(t) over w                                                        as "min(t)",
  max(t) over w                                                        as "max(t)"
from t
window w as (partition by class order by n range between unbounded preceding and current row)
order by class;

The GROUP BY clause has been removed; and the WINDOW definition clause has been added. And each aggregate function is invoked using the syntax defined by the fn_over_window rule. This is explained in the Semantics section within the Window function invocation—SQL syntax and semantics section. Notice that, in this example, the FILTER clause is used only for the aggregate function invocations that use numeric arguments.

Here is the result. Whitespace has been added manually around each of the rows that the FILTER clause identifies.

 class |   n    | avg(n)  | count(n) | min(n) | max(n) | sum(n) | t | count(expr) | min(t) | max(t)
-------+--------+---------+----------+--------+--------+--------+---+-------------+--------+--------
     1 |    100 |  100.00 |        1 |    100 |    100 |    100 | a |           1 | a      | a
     1 |    101 |  100.50 |        2 |    100 |    101 |    201 | b |           2 | a      | b

     1 |    102 |  100.50 |        2 |    100 |    101 |    201 | c |           3 | a      | c

     1 |    103 |  101.33 |        3 |    100 |    103 |    304 | d |           4 | a      | d
     1 | <null> |  101.33 |        3 |    100 |    103 |    304 | e |           4 | a      | e
     2 |    105 |  105.00 |        1 |    105 |    105 |    105 | f |           1 | f      | f
     2 |    106 |  105.50 |        2 |    105 |    106 |    211 | g |           2 | f      | g

     2 |    107 |  105.50 |        2 |    105 |    106 |    211 | h |           3 | f      | h

     2 |    108 |  106.33 |        3 |    105 |    108 |    319 | i |           4 | f      | i
     2 |    109 |  107.00 |        4 |    105 |    109 |    428 | j |           5 | f      | j

The choice, range between unbounded preceding and current row, for the frame_clause, was made because its effect is easy to understand and this lets you easily predict the values that the aggregate functions should produce for each successive row within the specified window.

Notice that the FILTER clause has no effect on the number of rows that the subquery defines. Rather, it determines only what rows within the window are considered by the particular aggregate function whose invocation is governed by that clause. The values shown in the columns "avg(n)", "count(n)", "min(n)", "max(n)", and "sum(n)" are the same for the filtered-out row with "n = 102" as they are for the previous row in the sorting order with "n = 101". The same holds for the filtered-out row with "n = 107" and the previous row with "n = 106".

The section Using the aggregate function avg() to compute a moving average shows how avg(), invoked with the OVER syntax, can be used to this effect.