For loop update better alternative - sql

Question

In Oracle 11g, I am using the following in a procedure.. can someone please provide a better solution to achieve the same results.
FOR REC IN
(SELECT E.EMP FROM EMPLOYEE E
JOIN
COMPANY C ON E.EMP=C.EMP
WHERE C.FLAG='Y')
LOOP
UPDATE EMPLOYEE SET FLAG='Y' WHERE EMP=REC.EMP;
END LOOP;
Is there a more efficient/better way to do this? I feel as if this method will run one update statement for each record found (Please correct me if I am wrong).
Here's the is actual code in full:
create or replace
PROCEDURE ACTION_MSC AS
BEGIN
-- ALL MIGRATED CONTACTS, CANDIDATES, COMPANIES, JOBS
-- ALL MIGRATED CANDIDATES, CONTACTS
FOR REC IN (SELECT DISTINCT AC.PEOPLE_HEX
FROM ACTION AC JOIN PEOPLE P ON AC.PEOPLE_HEX=P.PEOPLE_HEX
WHERE P.TO_MIGRATE='Y')
LOOP
UPDATE ACTION SET TO_MIGRATE='Y' WHERE PEOPLE_HEX=REC.PEOPLE_HEX;
END LOOP;
-- ALL MIGRATED COMPANIES
FOR REC IN (SELECT DISTINCT AC.COMPANY_HEX
FROM ACTION AC JOIN COMPANY CM ON AC.COMPANY_HEX=CM.COMPANY_HEX
WHERE CM.TO_MIGRATE='Y')
LOOP
UPDATE ACTION SET TO_MIGRATE='Y' WHERE COMPANY_HEX=REC.COMPANY_HEX;
END LOOP;
-- ALL MIGRATED JOBS
FOR REC IN (SELECT DISTINCT AC.JOB_HEX
FROM ACTION AC JOIN "JOB" J ON AC.JOB_HEX=J.JOB_HEX
WHERE J.TO_MIGRATE='Y')
LOOP
UPDATE ACTION SET TO_MIGRATE='Y' WHERE JOB_HEX=REC.JOB_HEX;
END LOOP;
COMMIT;
END ACTION_MSC;

Answer 1

You're right, it will do one update for each record found. Looks like you could just do:
UPDATE EMPLOYEE SET FLAG = 'Y'
WHERE EMP IN (SELECT EMP FROM COMPANY WHERE FLAG = 'Y')
AND FLAG != 'Y';
A single update will generally be faster and more efficient than multiple individual row updates in a loop; see this answer for another example. Apart from anything else, you're reducing the number of context switches between PL/SQL and SQL, which add up if you have a lot of rows. You could always benchmark this with your own data, of course.
I've added a check of the current flag state so you don't do a pointless update with no chamges.
It's fairly easy to compare the approaches to see that a single update is faster than one in a loop; with some contrived data:
create table people (id number, people_hex varchar2(16), to_migrate varchar2(1));
insert into people (id, people_hex, to_migrate)
select level, to_char(level - 1, 'xx'), 'Y'
from dual
connect by level <= 100;
create table action (id number, people_hex varchar2(16), to_migrate varchar2(1));
insert into action (id, people_hex, to_migrate)
select level, to_char(mod(level, 200), 'xx'), 'N'
from dual
connect by level <= 500000;
All of these will update half the rows in the action table. Updating in a loop:
begin
for rec in (select distinct ac.people_hex
from action ac join people p on ac.people_hex=p.people_hex
where p.to_migrate='Y')
loop
update action set to_migrate='Y' where people_hex=rec.people_hex;
end loop;
end;
/
Elapsed: 00:00:10.87
Single update (after rollback; I've left this in a block to mimic your procedure):
begin
update action set to_migrate = 'Y'
where people_hex in (select people_hex from people where to_migrate = 'Y');
end;
/
Elapsed: 00:00:07.14
Merge (after rollback):
begin
merge into action a
using (select people_hex, to_migrate from people where to_migrate = 'Y') p
on (a.people_hex = p.people_hex)
when matched then update set a.to_migrate = p.to_migrate;
end;
/
Elapsed: 00:00:07.00
There's some variation from repeated runs, particularly that update and merge are usually pretty close but sometimes swap which is faster in my environment; but both are always significantly faster than updating in a loop. You can repeat this in your own environment and with your own data spread and volumes, and you should if performance is that critical; but a single update is going to be faster than the loop. Whether you use update or merge isn't likely to make much difference.