Log row changes in PostgreSQL - Any advice/improvement is welcome ;-)

postgresql_changes_tracking.sql

/**
 * Creates a "logging" schema with an "history" table where are stored records in JSON format
 * 
 * Requires PostgreSQL >= 9.3 since data is stored in JSON format
 *
 * Credits: http://www.cybertec.at/2013/12/tracking-changes-in-postgresql/
 */

/* Create a schema dedicated to logs */
CREATE SCHEMA logging;

/* Create the table in which to store logs */
CREATE TABLE logging.history (
    id              serial,
    time            timestamptz DEFAULT CURRENT_TIMESTAMP,
    schema_name     text,
    table_name      text,
    user_name       text        DEFAULT current_user,
    operation       text,
    new_row         json,
    old_row         json
);

/* Create the function */
CREATE OR REPLACE FUNCTION logging.change_trigger() RETURNS trigger AS $$
    BEGIN
        IF    (TG_OP = 'INSERT') THEN
            INSERT INTO logging.history (table_name, schema_name, operation, new_row)
                VALUES (TG_RELNAME, TG_TABLE_SCHEMA, TG_OP, row_to_json(NEW));
            RETURN NEW;
        ELSIF (TG_OP = 'UPDATE') THEN
            INSERT INTO logging.history (table_name, schema_name, operation, new_row, old_row)
                VALUES (TG_RELNAME, TG_TABLE_SCHEMA, TG_OP, row_to_json(NEW), row_to_json(OLD));
            RETURN NEW;
        ELSIF (TG_OP = 'DELETE') THEN
            INSERT INTO logging.history (table_name, schema_name, operation, old_row)
                VALUES (TG_RELNAME, TG_TABLE_SCHEMA, TG_OP, row_to_json(OLD));
            RETURN OLD;
        END IF;
        RETURN NULL;
    END;
$$ LANGUAGE 'plpgsql' SECURITY DEFINER;

/* Create the triggers */
/* Single statement: change <schema_name> and <table_name> according to your needs */
/* 
DROP TRIGGER IF EXISTS table_change_trigger ON <schema_name>.<table_name>;
CREATE TRIGGER table_change_trigger 
    BEFORE INSERT OR UPDATE OR DELETE ON <schema_name>.<table_name>
    FOR EACH ROW EXECUTE PROCEDURE logging.change_trigger();
*/
/* Selects all tables in "public" schema and outputs the statements you need to execute for create the triggers */
SELECT 'DROP TRIGGER IF EXISTS table_change_trigger ON ' || schemaname || '.' || tablename || ';
    CREATE TRIGGER table_change_trigger 
    BEFORE INSERT OR UPDATE OR DELETE ON ' || schemaname || '.' || tablename || '
    FOR EACH ROW EXECUTE PROCEDURE logging.change_trigger();
' AS trigger_statements_to_execute
FROM pg_tables 
WHERE schemaname = 'public';

I've removed the check that prevents to insert a log if data hasn't changed after an UPDATE
because I've came up to an issue when there's a row with JSON column.
NEW <> OLD check fails with
ERROR: could not identify an equality operator for type json CONTEXT: PL/pgSQL

Do you know a way to properly check if NEW record is different to OLD record?

Should line 20 read new_row json? It looks like you are adding two fields each called old_row.

Yep, you're right!

I've tried this with an upsert (insert with on conflict do update) and the BEFORE trigger generated an insert and an update.
I've tried this with an AFTER trigger and it worked correctly (just an update or an insert).

Isn't it better to do all with AFTER triggers? In the original article (http://www.cybertec.at/tracking-changes-in-postgresql/) it is hinted in the comments that it would. I'm missing something?

Thanks! Is it possible to use the logging table to revert the table back to a pervious state?