PostgreSQL cannot begin/end transactions in PL/pgSQL

Question

I am seeking clarification of how to ensure an atomic transaction in a plpgsql function, and where the isolation level is set for this particular change to the database.

In the plpgsql function shown below, I want to make sure that BOTH the deletion AND the insertion succeed. I am getting an error when I try to wrap them in a single transaction:

ERROR:  cannot begin/end transactions in PL/pgSQL 

What happens during execution of the function below if another user has added a default behavior for circumstances ('RAIN', 'NIGHT', '45MPH') after this function has deleted the custom row but before it has had a chance to insert the custom row? Is there an implicit transaction wrapping the insert and delete so that both are rolled back if another user has changed either of the rows referenced by this function? Can I set the isolation level for this function?

create function foo(v_weather varchar(10), v_timeofday varchar(10), v_speed varchar(10),    v_behavior varchar(10))    returns setof CUSTOMBEHAVIOR    as $body$    begin       -- run-time error if either of these lines is un-commented        -- start transaction ISOLATION LEVEL READ COMMITTED;       -- or, alternatively, set transaction ISOLATION LEVEL READ COMMITTED;        delete from CUSTOMBEHAVIOR        where weather = 'RAIN' and timeofday = 'NIGHT' and speed= '45MPH' ;        -- if there is no default behavior insert a custom behavior        if not exists         (select id from DEFAULTBEHAVIOR where a = 'RAIN' and b = 'NIGHT' and c= '45MPH') then          insert into CUSTOMBEHAVIOR            (weather, timeofday, speed, behavior)          values            (v_weather, v_timeofday, v_speed, v_behavior);       end if;        return QUERY       select * from CUSTOMBEHAVIOR where ...   ;        -- commit;    end    $body$  LANGUAGE plpgsql; 

Answer 1

A plpgsql function automatically runs inside a transaction. It all succeeds or it all fails. The manual:

Functions and trigger procedures are always executed within a transaction established by an outer query — they cannot start or commit that transaction, since there would be no context for them to execute in. However, a block containing an EXCEPTION clause effectively forms a subtransaction that can be rolled back without affecting the outer transaction. For more about that see Section 42.6.6.

So, if you need to, you can catch an exception that theoretically might occur (but is very unlikely).
Details on trapping errors in the manual.

Your function reviewed and simplified:

CREATE FUNCTION foo(v_weather text                   , v_timeofday text                   , v_speed text                   , v_behavior text)   RETURNS SETOF custombehavior   LANGUAGE plpgsql AS $func$ BEGIN  DELETE FROM custombehavior WHERE  weather = 'RAIN' AND    timeofday = 'NIGHT' AND    speed = '45MPH';  INSERT INTO custombehavior (weather, timeofday, speed, behavior) SELECT v_weather, v_timeofday, v_speed, v_behavior WHERE  NOT EXISTS (    SELECT FROM defaultbehavior    WHERE  a = 'RAIN'    AND    b = 'NIGHT'    AND    c = '45MPH'    );  RETURN QUERY SELECT * FROM custombehavior WHERE ... ;  END $func$; 

---

If you actually need to begin/end transactions like indicated in the title look to SQL procedures in Postgres 11 or later (CREATE PROCEDURE). See:

• In PostgreSQL, what is the difference between a “Stored Procedure” and other types of functions?

Answer 2

Update: after PostgreSQL version 11. you can control transaction inside Store Procedure.

===== Before Version 10:

START TRANSACTION; select foo() ; COMMIT; 

"Unfortunately Postgres has no stored procedures, so you always need to manage the transaction in the calling code" – a_horse_with_no_name

Transaction in an exception block - how?