Using dbms_job
, you're never going to have two instances of a job running at the same time. If I do something like this with an interval
that is 1 minute in the future and a procedure that takes 3 minutes to run
declare
l_jobno pls_integer;
begin
dbms_job.submit( l_jobno,
'begin proc_that_takes_3_minutes(); end;',
sysdate,
'sysdate + interval ''1'' minute' );
end;
/
the interval
expression is only evaluated after the prior run finishes. There won't be 3 instances of the job running simultaneously. The job will just run once every 4 minutes.
This is why you frequently hear people about jobs "drifting" over time-- a job is scheduled to run at 4 AM with an interval of "sysdate + 1" so every day it runs at a slightly later time. If the job takes 1 minute to run, after 2 months, it's now running just after 5 AM. And that's why you'd generally want your interval
to be something like "trunc(sysdate+1) + interval '4' hour" if you want to run every day at 4 AM.
If you want to provide additional protection in case, say, someone tries to run the procedure in a different session at the same time that the job is running, you can put a dbms_lock.request
call in the procedure that tries to acquire a user-defined lock before doing the actual work.
create or replace procedure what_the_job_does
as
l_status integer;
begin
l_status := dbms_lock.request( 'MY_LOCK_NAME' );
if( l_status in (1,4) )
then
<<do actual work>>
end if;
end;
dbms_job
or thedbms_scheduler
package?