在 Postgres 中使用触发器

Question

I have two tables A and B in Postgres DB.

Whenever I insert a new row or rows in table AI want to add those updates to table B. If table B is empty, all the news rows inserted in table A will be also added to table B. However, when new rows are inserted in table A and if there is already data in table B, I want to delete all those data from table B before performing insert of the new rows of table A into table B.

So basically, every time when new rows are inserted into table A, corresponding rows should also be added to table B but perform delete operation every time before inserting to table B so that there is no old rows in table B. Table B is intended to hold only newly added rows of table A.

Following is the Postgres script where I have used a trigger.

    CREATE TRIGGER test_one AFTER INSERT ON A
    FOR EACH ROW EXECUTE PROCEDURE addRows();

    CREATE OR REPLACE FUNCTION addRows() RETURNS TRIGGER AS $remove_table$
       BEGIN
          DELETE from B;
          INSERT INTO B(std_id, name) VALUES (new.col1, new.col2);
          RETURN NEW;
       END;
    $remove_table$ LANGUAGE plpgsql;

This trigger function does the job but not all the newly added rows in table A are preserved. I end up with only one row in table B that is the last row of newly inserted data in the table A. To reemphasize, if I insert 10 new rows into table A, table B is expected to have 10 rows added. However, only the 10th row is added into table B which is not the desired result.

Can anybody help me understand what's wrong?

Answer 1

The problem with the trigger is that it fires for each row you insert, not each 'collection' of rows. With, for example, a batch of 10 inserts, the trigger will fire 10 times always removing all data in B, leaving you with 1 row - the last to be inserted.

Why create this additional table and duplicate data? Could you not store an insert date/time or an 'insert batch number' in table A and either query from this or create a view on it?

Answer 2

Since Postgres 10, pseudo tables are supported through which the inserted set can be accessed in FOR EACH STATEMENT triggers.

Change your function to do an INSERT... SELECT... from such a pseudo table.

CREATE
 OR REPLACE FUNCTION addrows
                     ()
                     RETURNS TRIGGER
AS
$$
BEGIN
  DELETE FROM b;
  
  INSERT INTO b
              (std_id,
               name)
              SELECT col1,
                     col2
                     FROM inserted;
  RETURN NULL;
END;
$$
LANGUAGE plpgsql;

And when creating the trigger, make it aa FOR EACH STATEMENT trigger and add a reference for the pseudo table.

CREATE TRIGGER test_one
               AFTER INSERT ON a
               REFERENCING NEW TABLE AS inserted
               FOR EACH STATEMENT
               EXECUTE PROCEDURE addrows();

db<>fiddle