Question

SELECT 
        id
        FROM table_name tkn1,
        (SELECT 
            id, 
            ROWNUM rnum
         FROM table_name 
         WHERE 
            PROCS_DT is null
         order by PRTY desc, CRET_DT) result 
        WHERE  tkn1.id= result.id
        AND result.rnum <= 10 FOR UPDATE OF tkn1.id SKIP LOCKED

Here is my problem. 2 threads are accessing this query at the same time

Thread 1 – Executes select and locks 10 rows ordered by descending priority and created date. Next I would update the procs_dt as todays date from a separate query..

Thread 2 – Before update of procs_dt or commit happens from thread 1 , this thread executes this query. My requirement is that the next 10 unlocked rows must be handed over to thread 2. But what really happens is the same set of locked rows comes out of the inner query since procs_dt is still null and yet to be updated by thread 1 and since skip locked is given in the outer query, all those 10 rows are skipped and no records are returned for thread 2 to process

This ultimately defeats my multi threading requirement.

How to fix this query? I tried adding the skip locked to the inner query. But oracle 11g doesn allow it.

Experts please help. I am using oracle 11g

Answer 1

I’d go with something like this :
A cursor to select the rows in order for update, and use the LIMIT clause to get the first ten available.

create table gm_temp
as select rownum id, table_name obj_name, date '2011-01-01' + rownum create_date 
from all_tables where rownum < 500;

CREATE TYPE tab_number IS TABLE OF NUMBER;

DECLARE
  cursor c_table IS 
    SELECT id FROM gm_temp ORDER BY create_date DESC FOR UPDATE OF id SKIP LOCKED;
  t_table_src tab_number := tab_number();
BEGIN
  OPEN c_table;
  FETCH c_table BULK COLLECT INTO t_table_src LIMIT 10;
  CLOSE c_table;
  dbms_output.put_line(':'||t_table_src.count||':'||t_table_src(1));
END;

Actually, I’d firstly see whether processing ALL outstanding rows as a set would be better than multi-threading.

Then if I did decide that I needed some form of multi-threading, I’d look at pipelined functions with parallel enabled (assuming I was on Enterprise Edition).