Postgresql function: get id of updated or inserted row - postgresql

Question

I have this function in my postgresql database that update row if exist or insert new one if it doesn't exist:
CREATE OR REPLACE FUNCTION insert_or_update(val1 integer, val2 integer) RETURNS VOID AS $$
DECLARE
BEGIN
UPDATE my_table SET col2 = val2 WHERE col1 = val1;
IF NOT FOUND THEN
INSERT INTO my_table (col2) values ( val2 );
END IF;
END;
$$ LANGUAGE 'plpgsql';
For now it's working perfect but I want to get the id of row if updated or inserted.
How can I do it?

Answer 1

Your function is declared as returns void so it can't return anything.
Assuming col1 is the primary key and is also defined as a serial, you can do something like this:
CREATE OR REPLACE FUNCTION insert_or_update(val1 integer, val2 integer)
RETURNS int
AS $$
DECLARE
l_id integer;
BEGIN
l_id := val1; -- initialize the local variable.
UPDATE my_table
SET col2 = val2
WHERE col1 = val1; -- !! IMPORTANT: this assumes col1 is unique !!
IF NOT FOUND THEN
INSERT INTO my_table (col2) values ( val2 )
RETURNING col1 -- this makes the generated value available
into l_id; -- and this stores it in the local variable
END IF;
return l_id; -- return whichever was used.
END;
$$ LANGUAGE plpgsql;
I changed four things compared to your function:
the function is declared as returns integer in order to be able to return something
you need a variable where you can store the returned value from the insert statement
and finally the generated value needs to be returned:
The language name is an identifier, so it must not be quoted using single quotes.
If you want to distinguish between an update or an insert from the caller, you could initialize l_id to null. In that case the function will return null if an update occurred and some value otherwise.

Answer 2

You can get the LastInsert ID using the method CURVAL(SEQUENCE_NAME_OF_TABLE).
But the best way is always to use the INSERT or UPDATE queries with RETURNING Clause.
CREATE OR REPLACE FUNCTION insert_or_update(val1 integer, val2 integer) RETURNS VOID AS $$
DECLARE
BEGIN
UPDATE my_table SET col2 = val2 WHERE col1 = val1 RETURNING col1;
IF NOT FOUND THEN
INSERT INTO my_table (col2) values ( val2 ) RETURNING col1;
END IF;
END;
$$ LANGUAGE 'plpgsql';
You can refer the following examples:
Insert Command - Last Example
Postgres with RETURNING clause
Note: In your UPDATE query, your WHERE clause is col1=val1. I assume that Val1 will be unique value, else multiple records will be updated. Hope you know that. And I assume col1 is your Primary Key like ID or so.

Answer 3

The PostgreSQL wiki's entry on UPSERT states that INSERT ... ON CONFLICT UPDATE will be added to PostgreSQL 9.5. This will allow you to more directly express the operation you desire without resorting to a stored procedure and/or introducing race conditions.
This operation is otherwise surprisingly tricky to express in earlier PostgreSQL versions without the risk of database corruption and/or a race condition. The code fragments posted so far all contain an error in that if two callers happen to want to upsert the same nonexistent row, the initial UPDATE will update zero rows and then they will both attempt an INSERT, one of which will fail. It should at least fail safe, aborting the query and any transaction in progress.
The PostgreSQL documentation on INSERT (search on that page for the text "Attempt to insert a new stock item along with the quantity of stock") shows how to do it safely and correctly on PostgreSQL 9.4 and earlier. Of particular note is that it tries the INSERT first to avoid any races on that front, and if that fails, does an UPDATE of the row it now knows exists. It uses a SAVEPOINT to ensure that a failed INSERT does not abort the transaction.