Question

I realize there are is no TRIGGER ON SELECT in Postgres. Given a table like so

CREATE TABLE t (
    a INTEGER PRIMARY KEY, 
    b TEXT, 
    entered_by INTEGER, 
    qry_count INTEGER
);

I want to increment “qry_count” for every SELECT for every “entered_by”, essentially keeping track of how many times any record is queried for every “enterer”. For example,

SELECT * a, b FROM t WHERE <condition>;

might return “n” rows entered by different enterers. For each enterer, I want to qry_count++. Pseudocode ahead

FOR EVERY entered_by IN SELECT
    UPDATE t
    SET qry_count = qry_count + 1
    WHERE entered_by = <entered_by>

I could do this most easily in my application, but I am wondering if doing this in the database itself might be best. I found an example of where I think I want to go, but it is for PL/SQL. What is the best way to accomplish this with Pg?

Update: In Perl I would do it like so

$sth_sel = $dbh->prepare( .. complicated SELECT includes "entered_by" ..);
$sth_upd = $dbh->prepare("UPDATE t SET qry_count = qry_count + 1 WHERE entered_by = ?");

$sth_sel->execute( .. bind params ..);
while (my $r = $sth_sel->fetchrow_arrayref) {
    my $entered_by = $r->[ 7 ]; # or whatever
    $sth_upd->execute($entered_by);

    .. do other things with $sth_sel, perhaps build a JSON obj to return ..
}

That does seem the easiest, but it would be nice to build this functionality right as a core part of the data schema.

UPDATE2: Most examples (including suggestions below) rely on creating a PL/PgSQL function. The problem with this approach is that my query is hard-coded in the function. Even if it can take input params, it is still a pre-declared query. This means I have to create a separate function for each query. In reality, in my application I construct the queries dynamically based on what is requested by the user (via the web). The requested columns can change, the supplied params can change. I think I am looking for a SQL analogous to the Perl pseudocode above without a predeclared function (SQL pseudocode below)

BEGIN
    FOR row IN
        SELECT to my hearts content
        FROM whatever tables I want JOINed horrendously
        WHERE any arbitrary param
    LOOP
        eb := row[entered_by]
        UPDATE t SET qry_count = qry_count + 1 WHERE entered_by = eb
        RETURN NEXT row;
    END LOOP;
    RETURN;
END

Hope this makes my objective clearer.

Answer 1

Notice the default value for the qry_count column:

CREATE TABLE t (
    a INTEGER PRIMARY KEY, 
    b TEXT, 
    entered_by INTEGER, 
    qry_count INTEGER default 0
);

create function select_and_update(parameter text)
returns setof t as $$
    update t
    set qry_count = qry_count + 1
    from (
        select a
        from t
        where b = $1
        ) s
    where t.a = s.a
    ;
    select *
    from t
    where b = $1
    ;
$$ language sql;

Now query the table using the above function:

select * from select_and_update('a');

Update according to comment:

You can build it dynamically and in instead of a function just wrap the sql code, whatever it is, in a transaction. No need for cursors.

begin;
    update t
    set qry_count = qry_count + 1
    from (
        select a
        from t
        where b = 'a'
        ) s
    where t.a = s.a
    ;
    select *
    from t
    where b = 'a'
    ;
commit;