Write data using utl_file

Question

hi , i am using oracle 10g.
i am writing this job which runs every day and writes into file.
first time i has to put upto 700,000 records into file and from next run it will be around 5000.
please let me know what ever i am doing is correct or not. i never used utl_file , since this is my first time using utl_file i really need help from you guys.
CREATE OR REPLACE PROCEDURE pr_cpe_dashboard
IS
   CURSOR c1
   IS
      SELECT /*+ parallel ( qdm 8) index(sdm INDX_RPRE_MART_SITE_QREV) index(odm INDX_RPRE_MART_ORD_QSITE) */
                qdm.quote_id
             || '||'
             || qdm.quote_revision
             || '||'
             || qdm.quote_status
             || '||'
             || qdm.last_modified_date
             || '||'
             || qdm.billing_method
             || '||'
             || qdm.quote_total
             || '||'
             || CASE
                   WHEN sdm.project_number IS NULL
                      THEN 'NULL'
                   ELSE sdm.project_number
                END
             || '||'
             || CASE
                   WHEN odm.order_number IS NULL
                      THEN 'NULL'
                   ELSE odm.order_number
                END
             || '||'
             || CASE
                   WHEN odm.order_type IS NULL
                      THEN 'NULL'
                   ELSE odm.order_type
                END
             || '||'
             || CASE
                   WHEN odm.release_timestamp IS NULL
                      THEN 'NULL'
                   ELSE TO_CHAR (odm.release_timestamp, 'mm/dd/yyyy hh:mi:ss')
                END
             || '||'
             || CASE
                   WHEN sdm.account_name IS NULL
                      THEN 'NULL'
                   ELSE sdm.account_name
                END
             || '||'
             || qdm.nasp_id
        FROM r_premisys_quote_detail_mart qdm,
             r_premisys_site_detail_mart sdm,
             r_premisys_order_detail_mart odm
       WHERE qdm.quote_id = sdm.quote_id(+)
         AND qdm.quote_revision = sdm.quote_revision(+)
         AND qdm.quote_id = odm.quote_id(+)
         AND qdm.quote_revision = odm.quote_revision(+)
         AND qdm.last_modified_date >= (SELECT last_used_date
                                          FROM job_audit_date);
   output_file   UTL_FILE.file_type;
   l_dir         VARCHAR2 (10)      := 'c:/orders';
   l_filename    VARCHAR2 (25)      := 'cpe.txt';
   v_array       t_array            := t_array ();
BEGIN
   output_file := UTL_FILE.fopen (l_dir, l_filename, 'W');
   OPEN c1;
   LOOP
      FETCH employees_cur
      BULK COLLECT INTO v_array LIMIT 1000;
      FOR i IN 1 .. v_array.COUNT
      LOOP
         UTL_FILE.put_line (output_file, v_array (i));
      END LOOP;
      EXIT WHEN v_array.COUNT = 0;
   END LOOP;
   UTL_FILE.fclose (output_file);
END pr_cpe_dashboard;

Answer 1

A couple of things.
This is wrong:
l_dir VARCHAR2 (10) := 'c:/orders';It must be the name of a directory object, created via the "+CREATE DIRECTORY+" command. Dir objects are (database) aliases for physical paths. Like any other db object, it provides a security layer that allows one to control access to it. You do not want any and all db sessions to access the root drive on that server.
Not necessary:
v_array t_array := t_array ();You are calling the constructor to create an empty array. Not needed as the bulk collect does that for you. Simply define the variable - the bulk collect takes the responsibility for initialising and populating it.
Review:
parallel ( qdm 8) Personally I dislike such hints, hints forcing indexes and so on. Reason is that the developer is second guessing the CBO. Yes, you may have the right values for the development db and its data set. It may even work for production for a while. But production is very seldom static. Process loads varies. Data volumes increases. Even the database model changes (e.g. new columns, new indexes, etc). H/w changes (e.g. more CPUs, more memory, etc).
By second guessing the CBO, the developer makes it very hard for the DBA and CBO properly manage performance and scalability on the server.
Also, keep in mind that UTL_FILE (and PL/SQL code) is a serialised process (only exception is specially crafted pipeline table functions). So despite using/requesting 8 parallel query processes (PQ) in the hint, a single PL process has to write that data to file. Be sure that you identify the appropriate bottleneck when dealing with I/O and wanting to use PQ to address it.