10g: delay for collecting results from parallel pipelined table functions
When parallel pipelined table functions are properly started and generate output record, there is a delay for the consuming main thread to gather these records.
This delay is huge compared with the run-time of the worker threads.
For my application it goes like this:
main thread timing efforts to start worker and collect their results:
[10:50:33-*10:50:49*]:JOMA: create (master): 015.93 sec (#66356 records, #4165/sec)
worker threads:
[10:50:34-*10:50:39*]:JOMA: create (slave) : 005.24 sec (#2449 EDRs, #467/sec, #0 errored / #6430 EBTMs, #1227/sec, #0 errored) - bulk #1 / sid #816
[10:50:34-*10:50:39*]:JOMA: create (slave) : 005.56 sec (#2543 EDRs, #457/sec, #0 errored / #6792 EBTMs, #1221/sec, #0 errored) - bulk #1 / sid #718
[10:50:34-*10:50:39*]:JOMA: create (slave) : 005.69 sec (#2610 EDRs, #459/sec, #0 errored / #6950 EBTMs, #1221/sec, #0 errored) - bulk #1 / sid #614
[10:50:34-*10:50:39*]:JOMA: create (slave) : 005.55 sec (#2548 EDRs, #459/sec, #0 errored / #6744 EBTMs, #1216/sec, #0 errored) - bulk #1 / sid #590
[10:50:34-*10:50:39*]:JOMA: create (slave) : 005.33 sec (#2461 EDRs, #462/sec, #0 errored / #6504 EBTMs, #1220/sec, #0 errored) - bulk #1 / sid #508
You can see, the worker threads are all started at the same time and terminating at the same time: 10:50:34-10:50:*39*.
But the main thread just invoking them and saving their results into a collection has finished at 10:50:*49*.
Why does it need #10 sec more just to save the data?
Here's a sample sqlplus script to demonstrate this:
--------------------------- snip -------------------------------------------------------
set serveroutput on;
drop table perf_data;
drop table test_table;
drop table tmp_test_table;
drop type ton_t;
drop type test_list;
drop type test_obj;
create table perf_data
sid number,
t1 timestamp with time zone,
t2 timestamp with time zone,
client varchar2(256)
create table test_table
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create global temporary table tmp_test_table
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_obj as object(
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_list as table of test_obj;
create or replace type ton_t as table of number;
create or replace package test_pkg
as
type test_rec is record (
a number(19,0),
b timestamp with time zone,
c varchar2(256)
type test_tab is table of test_rec;
type test_cur is ref cursor return test_rec;
function TZDeltaToMilliseconds(
t1 in timestamp with time zone,
t2 in timestamp with time zone)
return pls_integer;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a));
end;
create or replace package body test_pkg
as
* Calculate timestamp with timezone difference
* in milliseconds
function TZDeltaToMilliseconds(
t1 in timestamp with time zone,
t2 in timestamp with time zone)
return pls_integer
is
begin
return (extract(hour from t2) - extract(hour from t1)) * 3600 * 1000
+ (extract(minute from t2) - extract(minute from t1)) * 60 * 1000
+ (extract(second from t2) - extract(second from t1)) * 1000;
end TZDeltaToMilliseconds;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a))
is
pragma autonomous_transaction;
sid number;
counter number(19,0) := 0;
myrec test_rec;
mytab test_tab;
mytab2 test_list := test_list();
t1 timestamp with time zone;
t2 timestamp with time zone;
begin
t1 := systimestamp;
select userenv('SID') into sid from dual;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): enter');
loop
fetch mycur into myRec;
exit when mycur%NOTFOUND;
mytab2.extend;
mytab2(mytab2.last) := test_obj(myRec.a, myRec.b, myRec.c);
end loop;
for i in mytab2.first..mytab2.last loop
-- attention: saves own SID in test_obj.a for indication to caller
-- how many sids have been involved
pipe row(test_obj(sid, mytab2(i).b, mytab2(i).c));
pipe row(test_obj(sid, mytab2(i).b, mytab2(i).c)); -- duplicate
pipe row(test_obj(sid, mytab2(i).b, mytab2(i).c)); -- duplicate once again
counter := counter + 1;
end loop;
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'slave');
commit;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): exit, piped #' || counter || ' records');
end;
end;
declare
myList test_list := test_list();
myList2 test_list := test_list();
sids ton_t := ton_t();
sid number;
t1 timestamp with time zone;
t2 timestamp with time zone;
procedure LogPerfTable
is
type ton is table of number;
type tot is table of timestamp with time zone;
type clients_t is table of varchar2(256);
sids ton;
t1s tot;
t2s tot;
clients clients_t;
deltaTime integer;
btsPerSecond number(19,0);
edrsPerSecond number(19,0);
begin
select sid, t1, t2, client bulk collect into sids, t1s, t2s, clients from perf_data order by client;
if clients.count > 0 then
for i in clients.FIRST .. clients.LAST loop
deltaTime := test_pkg.TZDeltaToMilliseconds(t1s(i), t2s(i));
if deltaTime = 0 then deltaTime := 1; end if;
dbms_output.put_line(
'[' || to_char(t1s(i), 'hh:mi:ss') ||
'-' || to_char(t2s(i), 'hh:mi:ss') ||
']:' ||
' client ' || clients(i) || ' / sid #' || sids(i)
end loop;
end if;
end LogPerfTable;
begin
select userenv('SID') into sid from dual;
for i in 1..200000 loop
myList.extend; myList(myList.last) := test_obj(i, sysdate, to_char(i+2));
end loop;
-- save into the real table
insert into test_table select * from table(cast (myList as test_list));
-- save into the tmp table
insert into tmp_test_table select * from table(cast (myList as test_list));
dbms_output.put_line(chr(10) || '(1) copy ''mylist'' to ''mylist2'' by streaming via table function...');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from table(cast (myList as test_list)) tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(2) copy temporary ''tmp_test_table'' to ''mylist2'' by streaming via table function:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from tmp_test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(3) copy physical ''test_table'' to ''mylist2'' by streaming via table function:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
end;
--------------------------- snap -------------------------------------------------------
best regards,
Frank
Hello
I think the delay you are seeing is down to choosing the partitioning method as HASH. When you specify anything other than ANY, an additional buffer sort is included in the execution plan...
create or replace package test_pkg
as
type test_rec is record (
a number(19,0),
b timestamp with time zone,
c varchar2(256)
type test_tab is table of test_rec;
type test_cur is ref cursor return test_rec;
function TZDeltaToMilliseconds(
t1 in timestamp with time zone,
t2 in timestamp with time zone)
return pls_integer;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a));
function TF_Any(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by ANY);
end;
create or replace package body test_pkg
as
* Calculate timestamp with timezone difference
* in milliseconds
function TZDeltaToMilliseconds(
t1 in timestamp with time zone,
t2 in timestamp with time zone)
return pls_integer
is
begin
return (extract(hour from t2) - extract(hour from t1)) * 3600 * 1000
+ (extract(minute from t2) - extract(minute from t1)) * 60 * 1000
+ (extract(second from t2) - extract(second from t1)) * 1000;
end TZDeltaToMilliseconds;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a))
is
pragma autonomous_transaction;
sid number;
counter number(19,0) := 0;
myrec test_rec;
t1 timestamp with time zone;
t2 timestamp with time zone;
begin
t1 := systimestamp;
select userenv('SID') into sid from dual;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): enter');
loop
fetch mycur into myRec;
exit when mycur%NOTFOUND;
-- attention: saves own SID in test_obj.a for indication to caller
-- how many sids have been involved
pipe row(test_obj(sid, myRec.b, myRec.c));
pipe row(test_obj(sid, myRec.b, myRec.c)); -- duplicate
pipe row(test_obj(sid, myRec.b, myRec.c)); -- duplicate once again
counter := counter + 1;
end loop;
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'slave');
commit;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): exit, piped #' || counter || ' records');
end;
function TF_any(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by ANY)
is
pragma autonomous_transaction;
sid number;
counter number(19,0) := 0;
myrec test_rec;
t1 timestamp with time zone;
t2 timestamp with time zone;
begin
t1 := systimestamp;
select userenv('SID') into sid from dual;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): enter');
loop
fetch mycur into myRec;
exit when mycur%NOTFOUND;
-- attention: saves own SID in test_obj.a for indication to caller
-- how many sids have been involved
pipe row(test_obj(sid, myRec.b, myRec.c));
pipe row(test_obj(sid, myRec.b, myRec.c)); -- duplicate
pipe row(test_obj(sid, myRec.b, myRec.c)); -- duplicate once again
counter := counter + 1;
end loop;
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'slave');
commit;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): exit, piped #' || counter || ' records');
end;
end;
explain plan for
select /*+ first_rows */ test_obj(a, b, c)
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from test_table tab)));
select * from table(dbms_xplan.display);
Plan hash value: 1037943675
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | TQ |IN-OUT| PQ Distrib |
| 0 | SELECT STATEMENT | | 8168 | 3972K| 20 (0)| 00:00:01 | | | |
| 1 | PX COORDINATOR | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10001 | 8168 | 3972K| 20 (0)| 00:00:01 | Q1,01 | P->S | QC (RAND) |
| 3 | BUFFER SORT | | 8168 | 3972K| | | Q1,01 | PCWP | |
| 4 | VIEW | | 8168 | 3972K| 20 (0)| 00:00:01 | Q1,01 | PCWP | |
| 5 | COLLECTION ITERATOR PICKLER FETCH| TF | | | | | Q1,01 | PCWP | |
| 6 | PX RECEIVE | | 931K| 140M| 136 (2)| 00:00:02 | Q1,01 | PCWP | |
| 7 | PX SEND HASH | :TQ10000 | 931K| 140M| 136 (2)| 00:00:02 | Q1,00 | P->P | HASH |
| 8 | PX BLOCK ITERATOR | | 931K| 140M| 136 (2)| 00:00:02 | Q1,00 | PCWC | |
| 9 | TABLE ACCESS FULL | TEST_TABLE | 931K| 140M| 136 (2)| 00:00:02 | Q1,00 | PCWP | |
Note
- dynamic sampling used for this statement
explain plan for
select /*+ first_rows */ test_obj(a, b, c)
from table(test_pkg.TF_Any(CURSOR(select /*+ parallel(tab,5) */ * from test_table tab)));
select * from table(dbms_xplan.display);
Plan hash value: 4097140875
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | TQ |IN-OUT| PQ Distrib |
| 0 | SELECT STATEMENT | | 8168 | 3972K| 20 (0)| 00:00:01 | | | |
| 1 | PX COORDINATOR | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10000 | 8168 | 3972K| 20 (0)| 00:00:01 | Q1,00 | P->S | QC (RAND) |
| 3 | VIEW | | 8168 | 3972K| 20 (0)| 00:00:01 | Q1,00 | PCWP | |
| 4 | COLLECTION ITERATOR PICKLER FETCH| TF_ANY | | | | | Q1,00 | PCWP | |
| 5 | PX BLOCK ITERATOR | | 931K| 140M| 136 (2)| 00:00:02 | Q1,00 | PCWC | |
| 6 | TABLE ACCESS FULL | TEST_TABLE | 931K| 140M| 136 (2)| 00:00:02 | Q1,00 | PCWP | |
Note
- dynamic sampling used for this statementI posted about it here a few years ago and I more recently posted a question on Asktom. Unfortunately Tom was not able to find a technical reason for it to be there so I'm still a little in the dark as to why it is needed. The original question I posted is here:
Pipelined function partition by hash has extra sort#
I ran your tests with HASH vs ANY and the results are in line with the observations above....
declare
myList test_list := test_list();
myList2 test_list := test_list();
sids ton_t := ton_t();
sid number;
t1 timestamp with time zone;
t2 timestamp with time zone;
procedure LogPerfTable
is
type ton is table of number;
type tot is table of timestamp with time zone;
type clients_t is table of varchar2(256);
sids ton;
t1s tot;
t2s tot;
clients clients_t;
deltaTime integer;
btsPerSecond number(19,0);
edrsPerSecond number(19,0);
begin
select sid, t1, t2, client bulk collect into sids, t1s, t2s, clients from perf_data order by client;
if clients.count > 0 then
for i in clients.FIRST .. clients.LAST loop
deltaTime := test_pkg.TZDeltaToMilliseconds(t1s(i), t2s(i));
if deltaTime = 0 then deltaTime := 1; end if;
dbms_output.put_line(
'[' || to_char(t1s(i), 'hh:mi:ss') ||
'-' || to_char(t2s(i), 'hh:mi:ss') ||
']:' ||
' client ' || clients(i) || ' / sid #' || sids(i)
end loop;
end if;
end LogPerfTable;
begin
select userenv('SID') into sid from dual;
for i in 1..200000 loop
myList.extend; myList(myList.last) := test_obj(i, sysdate, to_char(i+2));
end loop;
-- save into the real table
insert into test_table select * from table(cast (myList as test_list));
-- save into the tmp table
insert into tmp_test_table select * from table(cast (myList as test_list));
dbms_output.put_line(chr(10) || '(1) copy ''mylist'' to ''mylist2'' by streaming via table function...');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from table(cast (myList as test_list)) tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(2) copy temporary ''tmp_test_table'' to ''mylist2'' by streaming via table function:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from tmp_test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(3) copy physical ''test_table'' to ''mylist2'' by streaming via table function:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,5) */ * from test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(4) copy temporary ''tmp_test_table'' to ''mylist2'' by streaming via table function ANY:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF_any(CURSOR(select /*+ parallel(tab,5) */ * from tmp_test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line(chr(10) || '(5) copy physical ''test_table'' to ''mylist2'' by streaming via table function using ANY:');
delete from perf_data;
t1 := systimestamp;
select /*+ first_rows */ test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF_any(CURSOR(select /*+ parallel(tab,5) */ * from test_table tab)));
t2 := systimestamp;
insert into perf_data (sid, t1, t2, client) values(sid, t1, t2, 'master');
LogPerfTable;
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
end;
(1) copy 'mylist' to 'mylist2' by streaming via table function...
test_pkg.TF( sid => '918' ): enter
test_pkg.TF( sid => '918' ): exit, piped #200000 records
[01:40:19-01:40:29]: client master / sid #918
[01:40:19-01:40:29]: client slave / sid #918
... saved #600000 records
(2) copy temporary 'tmp_test_table' to 'mylist2' by streaming via table function:
[01:40:31-01:40:36]: client master / sid #918
[01:40:31-01:40:32]: client slave / sid #659
[01:40:31-01:40:32]: client slave / sid #880
[01:40:31-01:40:32]: client slave / sid #1045
[01:40:31-01:40:32]: client slave / sid #963
[01:40:31-01:40:32]: client slave / sid #712
... saved #600000 records
(3) copy physical 'test_table' to 'mylist2' by streaming via table function:
[01:40:37-01:41:05]: client master / sid #918
[01:40:37-01:40:42]: client slave / sid #738
[01:40:37-01:40:42]: client slave / sid #568
[01:40:37-01:40:42]: client slave / sid #618
[01:40:37-01:40:42]: client slave / sid #659
[01:40:37-01:40:42]: client slave / sid #963
... saved #3000000 records
(4) copy temporary 'tmp_test_table' to 'mylist2' by streaming via table function ANY:
[01:41:12-01:41:16]: client master / sid #918
[01:41:12-01:41:16]: client slave / sid #712
[01:41:12-01:41:16]: client slave / sid #1045
[01:41:12-01:41:16]: client slave / sid #681
[01:41:12-01:41:16]: client slave / sid #754
[01:41:12-01:41:16]: client slave / sid #880
... saved #600000 records
(5) copy physical 'test_table' to 'mylist2' by streaming via table function using ANY:
[01:41:18-01:41:38]: client master / sid #918
[01:41:18-01:41:38]: client slave / sid #681
[01:41:18-01:41:38]: client slave / sid #712
[01:41:18-01:41:38]: client slave / sid #754
[01:41:18-01:41:37]: client slave / sid #880
[01:41:18-01:41:38]: client slave / sid #1045
... saved #3000000 recordsHTH
David
Similar Messages
-
Parallel pipelined table function, autonomous_transaction to global tmp tab
Hi,
i try to speed up my parallel pipelined table function and switch from pl/sql collection to global temporary table inside.
This requires to use PRAGMA AUTONOMOUS_TRANSACTION (and commit), because inserting into global temporary table (DML)
within select - for invoking the table function - is not allowed without.
As a consequence of commit it next requires to have on commit preserve rows for the global temporary table.
Now:
Inserts into the global temporary table are done - indicated by sql%rowcount.
But a select afterwards doesn't show any record anymore.
Here is a program to demonstrate it:
set serveroutput on;
drop type TestTableOfNumber_t;
create or replace type TestTableOfNumber_t is table of number;
drop type TestStatusList;
drop type TestStatusObj;
create or replace type TestStatusObj as object(
sid number,
ctr1 number,
ctr2 number,
ctr3 number
create or replace type TestStatusList is table of TestStatusObj;
drop table TestTmpTable;
create global temporary table TestTmpTable (
value number
) on commit preserve rows;
create or replace package test_pkg
as
type TestStatusRec is record (
sid number,
ctr1 number,
ctr2 number,
ctr3 number
type TestStatusTab is table of TestStatusRec;
function FillTmpTable(id in varchar2)
return TestStatusRec;
FUNCTION ptf (p_cursor IN sys_refcursor)
RETURN TestStatusList PIPELINED
PARALLEL_ENABLE(PARTITION p_cursor BY any);
end;
create or replace package body test_pkg
as
function FillTmpTable(id in varchar2)
return TestStatusRec
is
PRAGMA AUTONOMOUS_TRANSACTION;
result TestStatusRec;
sid number;
type ton is table of number;
tids TestTableOfNumber_t := TestTableOfNumber_t();
records number := 0;
begin
select userenv('SID') into sid from dual;
result.sid := sid;
delete from TestTmpTable;
for i in 1..100 loop
tids.extend;
tids(tids.last) := i;
end loop;
forall i in 1..tids.count
insert into TestTmpTable (value) values (tids(i));
-- get number of records inserted
records := sql%rowcount;
result.ctr1 := records;
-- retrieve again before commit
select count(*) into records from TestTmpTable;
result.ctr2 := records;
commit;
-- retrieve again after commit
select count(*) into records from TestTmpTable;
result.ctr3 := records;
return result;
end;
FUNCTION ptf (p_cursor IN sys_refcursor)
RETURN TestStatusList PIPELINED
PARALLEL_ENABLE(PARTITION p_cursor BY any)
IS
rec test_pkg.TestStatusRec;
value number;
sid number;
ctr integer := 0;
BEGIN
select userenv('SID') into sid from dual;
rec := FillTmpTable('IN PTF');
LOOP
FETCH p_cursor into value;
EXIT WHEN p_cursor%NOTFOUND;
ctr := ctr + 1;
END LOOP;
-- as a result i am only interested in the results of FillTmpTable():
PIPE ROW (TestStatusObj(rec.sid, rec.ctr1, rec.ctr2, rec.ctr3));
RETURN;
END;
end;
declare
tons TestTableOfNumber_t;
counts TestTableOfNumber_t;
status test_pkg.TestStatusRec;
statusList test_pkg.TestStatusTab;
begin
status := test_pkg.FillTmpTable('MAIN');
dbms_output.put_line('main thread:'
|| ' sid #' || status.sid
|| ' / #' || status.ctr1 || ' inserted '
|| ' / #' || status.ctr2 || ' before commit'
|| ' / #' || status.ctr3 || ' after commit');
select value bulk collect into tons from TestTmpTable;
select * bulk collect into statusList from TABLE(test_pkg.ptf(CURSOR(select /*+ parallel(tab,2) */ value from TestTmpTable tab)));
for i in 1..StatusList.count loop
dbms_output.put_line('worker thread #' || i || ':'
|| ' sid #' || statusList(i).sid
|| ' / #' || statusList(i).ctr1 || ' inserted '
|| ' / #' || statusList(i).ctr2 || ' before commit'
|| ' / #' || statusList(i).ctr3 || ' after commit');
end loop;
end;
/The output is:
main thread: sid #881 / #100 inserted / #100 before commit / #100 after commit
worker thread #1: sid #421 / #100 inserted / #0 before commit / #0 after commit
worker thread #2: sid #321 / #100 inserted / #0 before commit / #0 after commitThe 1st line is for the main thread invoking FillTmpTable().
The next #2 lines are for the worker threads of the parallel pipelined table function for invoking the same FillTmpTable().
For the main thread everything is as expected.
But for the worker threads the logs for before commit and after commit both give #0 for the number of available records in the global temporary table.
However all indicate #100 for the SQL insert
regards,
Frank
Edited by: user8704911 on Jul 7, 2011 10:13 AM
Edited by: user8704911 on Jul 7, 2011 10:20 AM
Edited by: user8704911 on Jul 7, 2011 10:27 AMSQL> select * from v$version;
BANNER
Oracle Database 11g Enterprise Edition Release 11.1.0.7.0 - 64bit Production
PL/SQL Release 11.1.0.7.0 - Production
CORE 11.1.0.7.0 Production
TNS for Linux: Version 11.1.0.7.0 - Production
NLSRTL Version 11.1.0.7.0 - Production
SQL> set serveroutput on;
SQL> drop type TestTableOfNumber;
drop type TestTableOfNumber
ERROR at line 1:
ORA-04043: object TESTTABLEOFNUMBER does not exist
SQL> /
drop type TestTableOfNumber
ERROR at line 1:
ORA-04043: object TESTTABLEOFNUMBER does not exist
SQL>
SQL> create or replace type TestTableOfNumber_t is table of number;
2 /
Type created.
SQL>
SQL> drop type TestStatusObj;
drop type TestStatusObj
ERROR at line 1:
ORA-04043: object TESTSTATUSOBJ does not exist
SQL> /
drop type TestStatusObj
ERROR at line 1:
ORA-04043: object TESTSTATUSOBJ does not exist
SQL>
SQL> create or replace type TestStatusObj as object(
2 sid number,
3 ctr1 number,
4 ctr2 number,
5 ctr3 number
6 );
7 /
Type created.
SQL>
SQL> drop type TestStatusList;
drop type TestStatusList
ERROR at line 1:
ORA-04043: object TESTSTATUSLIST does not exist
SQL> /
drop type TestStatusList
ERROR at line 1:
ORA-04043: object TESTSTATUSLIST does not exist
SQL>
SQL> create or replace type TestStatusList is table of TestStatusObj;
2 /
Type created.
SQL>
SQL> drop table TestTmpTable;
drop table TestTmpTable
ERROR at line 1:
ORA-00942: table or view does not exist
SQL> /
drop table TestTmpTable
ERROR at line 1:
ORA-00942: table or view does not exist
SQL>
SQL> create global temporary table TestTmpTable (
2 value number
3 ) on commit preserve rows;
Table created.
SQL> /
create global temporary table TestTmpTable (
ERROR at line 1:
ORA-00955: name is already used by an existing object
SQL>
SQL> create or replace package test_pkg
2 as
3
4 type TestStatusRec is record (
5 sid number,
6 ctr1 number,
7 ctr2 number,
8 ctr3 number
9 );
10
11 type TestStatusTab is table of TestStatusRec;
12
13 function FillTmpTable(id in varchar2)
14 return TestStatusRec;
15
16 FUNCTION ptf (p_cursor IN sys_refcursor)
17 RETURN TestStatusList PIPELINED
18 PARALLEL_ENABLE(PARTITION p_cursor BY any);
19
20 end;
21 /
Package created.
SQL>
SQL> create or replace package body test_pkg
2 as
3
4 function FillTmpTable(id in varchar2)
5 return TestStatusRec
6 is
7 PRAGMA AUTONOMOUS_TRANSACTION;
8
9 result TestStatusRec;
10
11 sid number;
12
13 type ton is table of number;
14 tids TestTableOfNumber_t := TestTableOfNumber_t();
15
16 records number := 0;
17 begin
18 select userenv('SID') into sid from dual;
19 result.sid := sid;
20
21 delete from TestTmpTable;
22
23 for i in 1..100 loop
24 tids.extend;
25 tids(tids.last) := i;
26 end loop;
27
28 forall i in 1..tids.count
29 insert into TestTmpTable (value) values (tids(i));
30
31 -- get number of records inserted
32 records := sql%rowcount;
33 result.ctr1 := records;
34
35 -- retrieve again before commit
36 select count(*) into records from TestTmpTable;
37 result.ctr2 := records;
38
39 commit;
40
41 -- retrieve again after commit
42 select count(*) into records from TestTmpTable;
43 result.ctr3 := records;
44
45 return result;
46 end;
47
48 FUNCTION ptf (p_cursor IN sys_refcursor)
49 RETURN TestStatusList PIPELINED
50 PARALLEL_ENABLE(PARTITION p_cursor BY any)
51 IS
52 rec test_pkg.TestStatusRec;
53 value number;
54 sid number;
55 ctr integer := 0;
56 BEGIN
57 select userenv('SID') into sid from dual;
58 rec := FillTmpTable('IN PTF');
59 LOOP
60 FETCH p_cursor into value;
61 EXIT WHEN p_cursor%NOTFOUND;
62 ctr := ctr + 1;
63 END LOOP;
64
65 -- as a result i am only interested in the results of FillTmpTable():
66 PIPE ROW (TestStatusObj(rec.sid, rec.ctr1, rec.ctr2, rec.ctr3));
67
68 RETURN;
69 END;
70 end;
71 /
Package body created.
SQL>
SQL> declare
2 tons TestTableOfNumber_t;
3 counts TestTableOfNumber_t;
4 status test_pkg.TestStatusRec;
5 statusList test_pkg.TestStatusTab;
6 begin
7 status := test_pkg.FillTmpTable('MAIN');
8 dbms_output.put_line('main thread:'
9 || ' sid #' || status.sid
10 || ' / #' || status.ctr1 || ' inserted '
11 || ' / #' || status.ctr2 || ' before commit'
12 || ' / #' || status.ctr3 || ' after commit');
13
14 select value bulk collect into tons from TestTmpTable;
15
16 select * bulk collect into statusList from TABLE(test_pkg.ptf(CURSOR(select /*+ parallel(tab,2
) */ value from TestTmpTable tab)));
17
18 for i in 1..StatusList.count loop
19 dbms_output.put_line('worker thread #' || i || ':'
20 || ' sid #' || statusList(i).sid
21 || ' / #' || statusList(i).ctr1 || ' inserted '
22 || ' / #' || statusList(i).ctr2 || ' before commit'
23 || ' / #' || statusList(i).ctr3 || ' after commit');
24 end loop;
25
26 end;
27 /
main thread: sid #1023 / #100 inserted / #100 before commit / #100 after commit
worker thread #1: sid #1045 / #100 inserted / #100 before commit / #100 after
commit
worker thread #2: sid #1019 / #100 inserted / #100 before commit / #100 after
commit
PL/SQL procedure successfully completed.
SQL> I am getting a different result.
Regards
Raj -
10g: parallel pipelined table func. using table(cast(SQL collect.))?
Hi,
i try to distribute SQL data objects - stored in a SQL data type TABLE OF <object-Type> - to multiple (parallel) instances of a table function,
by passing a CURSOR(...) to the table function, which selects from the SQL TABLE OF storage via "select * from TABLE(CAST(<storage> as <storage-type>)".
But oracle always only uses a single table function instance :-(
whatever hints i provide or setting i use for the parallel table function (parallel_enable ...)
Could it be, that this is due to the fact, that my data are not
globally available, but only in the main thread data?
Can someone confirm, that it's not possible to start multiple parallel table functions
for selecting on SQL data type TABLE OF <object>storages?
Here's an example sqlplus program to show the issue:
-------------------- snip ---------------------------------------------
set serveroutput on;
drop table test_table;
drop type ton_t;
drop type test_list;
drop type test_obj;
create table test_table
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_obj as object(
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_list as table of test_obj;
create or replace type ton_t as table of number;
create or replace package test_pkg
as
type test_rec is record (
a number(19,0),
b timestamp with time zone,
c varchar2(256)
type test_tab is table of test_rec;
type test_cur is ref cursor return test_rec;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a));
end;
create or replace package body test_pkg
as
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a))
is
sid number;
counter number(19,0) := 0;
myrec test_rec;
mytab test_tab;
mytab2 test_list := test_list();
begin
select userenv('SID') into sid from dual;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): enter');
loop
fetch mycur into myRec;
exit when mycur%NOTFOUND;
mytab2.extend;
mytab2(mytab2.last) := test_obj(myRec.a, myRec.b, myRec.c);
end loop;
for i in mytab2.first..mytab2.last loop
-- attention: saves own SID in test_obj.a for indication to caller
-- how many sids have been involved
pipe row(test_obj(sid, mytab2(i).b, mytab2(i).c));
counter := counter + 1;
end loop;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): exit, piped #' || counter || ' records');
end;
end;
declare
myList test_list := test_list();
myList2 test_list := test_list();
sids ton_t := ton_t();
begin
for i in 1..10000 loop
myList.extend; myList(myList.last) := test_obj(i, sysdate, to_char(i+2));
end loop;
-- save into the real table
insert into test_table select * from table(cast (myList as test_list));
dbms_output.put_line(chr(10) || 'copy ''mylist'' to ''mylist2'' by streaming via table function...');
select test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,10) */ * from table(cast (myList as test_list)) tab)));
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line('worker thread''s sid list:');
for i in sids.first..sids.last loop
dbms_output.put_line('sid #' || sids(i));
end loop;
dbms_output.put_line(chr(10) || 'copy physical ''test_table'' to ''mylist2'' by streaming via table function:');
select test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,10) */ * from test_table tab)));
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line('worker thread''s sid list:');
for i in sids.first..sids.last loop
dbms_output.put_line('sid #' || sids(i));
end loop;
end;
-------------------- snap ---------------------------------------------
Here's the output:
-------------------- snip ---------------------------------------------
copy 'mylist' to 'mylist2' by streaming via table function...
test_pkg.TF( sid => '98' ): enter
test_pkg.TF( sid => '98' ): exit, piped #10000 records
... saved #10000 records
worker thread's sid list:
sid #98 -- ONLY A SINGLE SID HERE!
copy physical 'test_table' to 'mylist2' by streaming via table function:
... saved #10000 records
worker thread's sid list:
sid #128 -- A LIST OF SIDS HERE!
sid #141
sid #85
sid #125
sid #254
sid #101
sid #124
sid #109
sid #142
sid #92
PL/SQL procedure successfully completed.
-------------------- snap ---------------------------------------------
I posted it to newsgroup comp.databases.oracle.server.
(summary: "10g: parallel pipelined table functions with cursor selecting from table(cast(SQL collection)) doesn't work ")
But i didn't get a response.
There i also wrote some background information about my application:
-------------------- snip ---------------------------------------------
My application has a #2 steps/stages data selection.
A 1st select for minimal context base data
- mainly to evaluate for due driving data records.
And a 2nd select for all the "real" data to process a context
(joining much more other tables here, which i don't want to do for non-due records).
So it's doing stage #1 select first, then stage #2 select - based on stage #1 results - next.
The first implementation of the application did the stage #1 select in the main session of the pl/sql code.
And for the stage #2 select there was done a dispatch to multiple parallel table functions (in multiple worker sessions) for the "real work".
That worked.
However there was a flaw:
Between records from stage #1 selection and records from stage #2 selection there is a 1:n relation (via key / foreign key relation).
Means, for #1 resulting record from stage #1 selection, there are #x records from stage #2 selection.
That forced me to use "cluster curStage2 by (theKey)".
Because the worker sessions need to evaluate the all-over status for a context of #1 record from stage #1 and #x records from stage #2
(so it needs to have #x records of stage #2 together).
This then resulted in delay for starting up the worker sessions (i didn't find a way to get rid of this).
So i wanted to shift the invocation of the worker sessions to the stage #1 selection.
Then i don't need the "cluster curStage2 by (theKey)" anymore!
But: i also need to do an update of the primary driving data!
So the stage #1 select is a 'select ... for update ...'.
But you can't use such in CURSOR for table functions (which i can understand, why it's not possible).
So i have to do my stage #1 selection in two steps:
1. 'select for update' by main session and collect result in SQL collection.
2. pass collected data to parallel table functions
And for 2. i recognized, that it doesn't start up multiple parallel table function instances.
As a work-around
- if it's just not possible to start multiple parallel pipelined table functions for dispatching from 'select * from TABLE(CAST(... as ...))' -
i need to select again on the base tables - driven by the SQL collection data.
But before i do so, i wanted to verify, if it's really not possible.
Maybe i just miss a special oracle hint or whatever you can get "out of another box" :-)
-------------------- snap ---------------------------------------------
- many thanks!
rgds,
FrankHi,
i try to distribute SQL data objects - stored in a SQL data type TABLE OF <object-Type> - to multiple (parallel) instances of a table function,
by passing a CURSOR(...) to the table function, which selects from the SQL TABLE OF storage via "select * from TABLE(CAST(<storage> as <storage-type>)".
But oracle always only uses a single table function instance :-(
whatever hints i provide or setting i use for the parallel table function (parallel_enable ...)
Could it be, that this is due to the fact, that my data are not
globally available, but only in the main thread data?
Can someone confirm, that it's not possible to start multiple parallel table functions
for selecting on SQL data type TABLE OF <object>storages?
Here's an example sqlplus program to show the issue:
-------------------- snip ---------------------------------------------
set serveroutput on;
drop table test_table;
drop type ton_t;
drop type test_list;
drop type test_obj;
create table test_table
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_obj as object(
a number(19,0),
b timestamp with time zone,
c varchar2(256)
create or replace type test_list as table of test_obj;
create or replace type ton_t as table of number;
create or replace package test_pkg
as
type test_rec is record (
a number(19,0),
b timestamp with time zone,
c varchar2(256)
type test_tab is table of test_rec;
type test_cur is ref cursor return test_rec;
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a));
end;
create or replace package body test_pkg
as
function TF(mycur test_cur)
return test_list pipelined
parallel_enable(partition mycur by hash(a))
is
sid number;
counter number(19,0) := 0;
myrec test_rec;
mytab test_tab;
mytab2 test_list := test_list();
begin
select userenv('SID') into sid from dual;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): enter');
loop
fetch mycur into myRec;
exit when mycur%NOTFOUND;
mytab2.extend;
mytab2(mytab2.last) := test_obj(myRec.a, myRec.b, myRec.c);
end loop;
for i in mytab2.first..mytab2.last loop
-- attention: saves own SID in test_obj.a for indication to caller
-- how many sids have been involved
pipe row(test_obj(sid, mytab2(i).b, mytab2(i).c));
counter := counter + 1;
end loop;
dbms_output.put_line('test_pkg.TF( sid => '''|| sid || ''' ): exit, piped #' || counter || ' records');
end;
end;
declare
myList test_list := test_list();
myList2 test_list := test_list();
sids ton_t := ton_t();
begin
for i in 1..10000 loop
myList.extend; myList(myList.last) := test_obj(i, sysdate, to_char(i+2));
end loop;
-- save into the real table
insert into test_table select * from table(cast (myList as test_list));
dbms_output.put_line(chr(10) || 'copy ''mylist'' to ''mylist2'' by streaming via table function...');
select test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,10) */ * from table(cast (myList as test_list)) tab)));
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line('worker thread''s sid list:');
for i in sids.first..sids.last loop
dbms_output.put_line('sid #' || sids(i));
end loop;
dbms_output.put_line(chr(10) || 'copy physical ''test_table'' to ''mylist2'' by streaming via table function:');
select test_obj(a, b, c) bulk collect into myList2
from table(test_pkg.TF(CURSOR(select /*+ parallel(tab,10) */ * from test_table tab)));
dbms_output.put_line('... saved #' || myList2.count || ' records');
select distinct(tab.a) bulk collect into sids from table(cast (myList2 as test_list)) tab;
dbms_output.put_line('worker thread''s sid list:');
for i in sids.first..sids.last loop
dbms_output.put_line('sid #' || sids(i));
end loop;
end;
-------------------- snap ---------------------------------------------
Here's the output:
-------------------- snip ---------------------------------------------
copy 'mylist' to 'mylist2' by streaming via table function...
test_pkg.TF( sid => '98' ): enter
test_pkg.TF( sid => '98' ): exit, piped #10000 records
... saved #10000 records
worker thread's sid list:
sid #98 -- ONLY A SINGLE SID HERE!
copy physical 'test_table' to 'mylist2' by streaming via table function:
... saved #10000 records
worker thread's sid list:
sid #128 -- A LIST OF SIDS HERE!
sid #141
sid #85
sid #125
sid #254
sid #101
sid #124
sid #109
sid #142
sid #92
PL/SQL procedure successfully completed.
-------------------- snap ---------------------------------------------
I posted it to newsgroup comp.databases.oracle.server.
(summary: "10g: parallel pipelined table functions with cursor selecting from table(cast(SQL collection)) doesn't work ")
But i didn't get a response.
There i also wrote some background information about my application:
-------------------- snip ---------------------------------------------
My application has a #2 steps/stages data selection.
A 1st select for minimal context base data
- mainly to evaluate for due driving data records.
And a 2nd select for all the "real" data to process a context
(joining much more other tables here, which i don't want to do for non-due records).
So it's doing stage #1 select first, then stage #2 select - based on stage #1 results - next.
The first implementation of the application did the stage #1 select in the main session of the pl/sql code.
And for the stage #2 select there was done a dispatch to multiple parallel table functions (in multiple worker sessions) for the "real work".
That worked.
However there was a flaw:
Between records from stage #1 selection and records from stage #2 selection there is a 1:n relation (via key / foreign key relation).
Means, for #1 resulting record from stage #1 selection, there are #x records from stage #2 selection.
That forced me to use "cluster curStage2 by (theKey)".
Because the worker sessions need to evaluate the all-over status for a context of #1 record from stage #1 and #x records from stage #2
(so it needs to have #x records of stage #2 together).
This then resulted in delay for starting up the worker sessions (i didn't find a way to get rid of this).
So i wanted to shift the invocation of the worker sessions to the stage #1 selection.
Then i don't need the "cluster curStage2 by (theKey)" anymore!
But: i also need to do an update of the primary driving data!
So the stage #1 select is a 'select ... for update ...'.
But you can't use such in CURSOR for table functions (which i can understand, why it's not possible).
So i have to do my stage #1 selection in two steps:
1. 'select for update' by main session and collect result in SQL collection.
2. pass collected data to parallel table functions
And for 2. i recognized, that it doesn't start up multiple parallel table function instances.
As a work-around
- if it's just not possible to start multiple parallel pipelined table functions for dispatching from 'select * from TABLE(CAST(... as ...))' -
i need to select again on the base tables - driven by the SQL collection data.
But before i do so, i wanted to verify, if it's really not possible.
Maybe i just miss a special oracle hint or whatever you can get "out of another box" :-)
-------------------- snap ---------------------------------------------
- many thanks!
rgds,
Frank -
10g: parallel pipelined table func - distributing DISTINCT data sets
Hi,
i want to distribute data records, selected from cursor, via parallel pipelined table function to multiple worker threads for processing and returning result records.
The tables, where i am selecting data from, are partitioned and subpartitioned.
All tables share the same partitioning/subpartitioning schema.
Each table has a column 'Subpartition_Key', which is hashed to a physical subpartition.
E.g. the Subpartition_Key ranges from 000...999, but we have only 10 physical subpartitions.
The select of records is done partition-wise - one partition after another (in bulks).
The parallel running worker threads select more data from other tables for their processing (2nd level select)
Now my goal is to distribute initial records to the worker threads in a way, that they operate on distinct subpartitions - to decouple the access to resources (for the 2nd level select)
But i cannot just use 'parallel_enable(partition curStage1 by hash(subpartition_key))' for the distribution.
hash(subpartition_key) (hashing A) does not match with the hashing B used to assign the physical subpartition for the INSERT into the tables.
Even when i remodel the hashing B, calculate some SubPartNo(subpartition_key) and use that for 'parallel_enable(partition curStage1 by hash(SubPartNo))' it doesn't work.
Also 'parallel_enable(partition curStage1 by range(SubPartNo))' doesn't help. The load distribution is unbalanced - some worker threads get data of one subpartition, some of multiple subpartitions, some are idle.
How can i distribute the records to the worker threads according a given subpartition-schema?
+[amendment:+
Actually the hashing for the parallel_enable is counterproductive here - it would be better to have some 'parallel_enable(partition curStage1 by SubPartNo)'.]
- many thanks!
best regards,
Frank
Edited by: user8704911 on Jan 12, 2012 2:51 AMHello
A couple of things to note. 1, when you use partition by hash(or range) on 10gr2 and above, there is an additional BUFFER SORT operation vs using partition by ANY. For small datasets this is not necessarily an issue, but the temp space used by this stage can be significant for larger data sets. So be sure to check temp space usage for this process or you could run into problems later.
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop | TQ |IN-OUT| PQ Distrib |
| 0 | SELECT STATEMENT | | 8168 | 1722K| 24 (0)| 00:00:01 | | | | | |
| 1 | PX COORDINATOR | | | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10001 | 8168 | 1722K| 24 (0)| 00:00:01 | | | Q1,01 | P->S | QC (RAND) |
| 3 |****BUFFER SORT**** | | 8168 | 1722K| | | | | Q1,01 | PCWP | |
| 4 | VIEW | | 8168 | 1722K| 24 (0)| 00:00:01 | | | Q1,01 | PCWP | |
| 5 | COLLECTION ITERATOR PICKLER FETCH| TF | | | | | | | Q1,01 | PCWP | |
| 6 | PX RECEIVE | | 100 | 4800 | 2 (0)| 00:00:01 | | | Q1,01 | PCWP | |
| 7 | PX SEND HASH | :TQ10000 | 100 | 4800 | 2 (0)| 00:00:01 | | | Q1,00 | P->P | HASH |
| 8 | PX BLOCK ITERATOR | | 100 | 4800 | 2 (0)| 00:00:01 | 1 | 10 | Q1,00 | PCWC | |
| 9 | TABLE ACCESS FULL | TEST_TAB | 100 | 4800 | 2 (0)| 00:00:01 | 1 | 20 | Q1,00 | PCWP | |
-----------------------------------------------------------------------------------------------------------------------------------------------It may be in this case that you can use clustering with partition by any to achieve your goal...
create or replace package test_pkg as
type Test_Tab_Rec_t is record (
Tracking_ID number(19),
Partition_Key date,
Subpartition_Key number(3),
sid number
type Test_Tab_Rec_Tab_t is table of Test_Tab_Rec_t;
type Test_Tab_Rec_Hash_t is table of Test_Tab_Rec_t index by binary_integer;
type Test_Tab_Rec_HashHash_t is table of Test_Tab_Rec_Hash_t index by binary_integer;
type Cur_t is ref cursor return Test_Tab_Rec_t;
procedure populate;
procedure report;
function tf(cur in Cur_t)
return test_list pipelined
parallel_enable(partition cur by hash(subpartition_key));
function tf_any(cur in Cur_t)
return test_list PIPELINED
CLUSTER cur BY (Subpartition_Key)
parallel_enable(partition cur by ANY);
end;
create or replace package body test_pkg as
procedure populate
is
Tracking_ID number(19) := 1;
Partition_Key date := current_timestamp;
Subpartition_Key number(3) := 1;
begin
dbms_output.put_line(chr(10) || 'populate data into Test_Tab...');
for Subpartition_Key in 0..99
loop
for ctr in 1..1
loop
insert into test_tab (tracking_id, partition_key, subpartition_key)
values (Tracking_ID, Partition_Key, Subpartition_Key);
Tracking_ID := Tracking_ID + 1;
end loop;
end loop;
dbms_output.put_line('...done (populate data into Test_Tab)');
end;
procedure report
is
recs Test_Tab_Rec_Tab_t;
begin
dbms_output.put_line(chr(10) || 'list data per partition/subpartition...');
for item in (select partition_name, subpartition_name from user_tab_subpartitions where table_name='TEST_TAB' order by partition_name, subpartition_name)
loop
dbms_output.put_line('partition/subpartition = ' || item.partition_name || '/' || item.subpartition_name || ':');
execute immediate 'select * from test_tab SUBPARTITION(' || item.subpartition_name || ')' bulk collect into recs;
if recs.count > 0
then
for i in recs.first..recs.last
loop
dbms_output.put_line('...' || recs(i).Tracking_ID || ', ' || recs(i).Partition_Key || ', ' || recs(i).Subpartition_Key);
end loop;
end if;
end loop;
dbms_output.put_line('... done (list data per partition/subpartition)');
end;
function tf(cur in Cur_t)
return test_list pipelined
parallel_enable(partition cur by hash(subpartition_key))
is
sid number;
input Test_Tab_Rec_t;
output test_object;
begin
select userenv('SID') into sid from dual;
loop
fetch cur into input;
exit when cur%notfound;
output := test_object(input.tracking_id, input.partition_key, input.subpartition_key,sid);
pipe row(output);
end loop;
end;
function tf_any(cur in Cur_t)
return test_list PIPELINED
CLUSTER cur BY (Subpartition_Key)
parallel_enable(partition cur by ANY)
is
sid number;
input Test_Tab_Rec_t;
output test_object;
begin
select userenv('SID') into sid from dual;
loop
fetch cur into input;
exit when cur%notfound;
output := test_object(input.tracking_id, input.partition_key, input.subpartition_key,sid);
pipe row(output);
end loop;
end;
end;
XXXX> with parts as (
2 select --+ materialize
3 data_object_id,
4 subobject_name
5 FROM
6 user_objects
7 WHERE
8 object_name = 'TEST_TAB'
9 and
10 object_type = 'TABLE SUBPARTITION'
11 )
12 SELECT
13 COUNT(*),
14 parts.subobject_name,
15 target.sid
16 FROM
17 parts,
18 test_tab tt,
19 test_tab_part_hash target
20 WHERE
21 tt.tracking_id = target.tracking_id
22 and
23 parts.data_object_id = DBMS_MView.PMarker(tt.rowid)
24 GROUP BY
25 parts.subobject_name,
26 target.sid
27 ORDER BY
28 target.sid,
29 parts.subobject_name
30 /
XXXX> INSERT INTO test_tab_part_hash select * from table(test_pkg.tf(CURSOR(select * from test_tab)))
2 /
100 rows created.
Elapsed: 00:00:00.14
XXXX>
XXXX> INSERT INTO test_tab_part_any_cluster select * from table(test_pkg.tf_any(CURSOR(select * from test_tab)))
2 /
100 rows created.
--using partition by hash
XXXX> with parts as (
2 select --+ materialize
3 data_object_id,
4 subobject_name
5 FROM
6 user_objects
7 WHERE
8 object_name = 'TEST_TAB'
9 and
10 object_type = 'TABLE SUBPARTITION'
11 )
12 SELECT
13 COUNT(*),
14 parts.subobject_name,
15 target.sid
16 FROM
17 parts,
18 test_tab tt,
19 test_tab_part_hash target
20 WHERE
21 tt.tracking_id = target.tracking_id
22 and
23 parts.data_object_id = DBMS_MView.PMarker(tt.rowid)
24 GROUP BY
25 parts.subobject_name,
26 target.sid
27 /
COUNT(*) SUBOBJECT_NAME SID
3 SYS_SUBP31 1272
1 SYS_SUBP32 1272
1 SYS_SUBP33 1272
3 SYS_SUBP34 1272
1 SYS_SUBP36 1272
1 SYS_SUBP37 1272
3 SYS_SUBP38 1272
1 SYS_SUBP39 1272
1 SYS_SUBP32 1280
2 SYS_SUBP33 1280
2 SYS_SUBP34 1280
1 SYS_SUBP35 1280
2 SYS_SUBP36 1280
1 SYS_SUBP37 1280
2 SYS_SUBP38 1280
1 SYS_SUBP40 1280
2 SYS_SUBP33 1283
2 SYS_SUBP34 1283
2 SYS_SUBP35 1283
2 SYS_SUBP36 1283
1 SYS_SUBP37 1283
1 SYS_SUBP38 1283
2 SYS_SUBP39 1283
1 SYS_SUBP40 1283
1 SYS_SUBP32 1298
1 SYS_SUBP34 1298
1 SYS_SUBP36 1298
2 SYS_SUBP37 1298
4 SYS_SUBP38 1298
2 SYS_SUBP40 1298
1 SYS_SUBP31 1313
1 SYS_SUBP33 1313
1 SYS_SUBP39 1313
1 SYS_SUBP40 1313
1 SYS_SUBP32 1314
1 SYS_SUBP35 1314
1 SYS_SUBP38 1314
1 SYS_SUBP40 1314
2 SYS_SUBP33 1381
1 SYS_SUBP34 1381
1 SYS_SUBP35 1381
3 SYS_SUBP36 1381
3 SYS_SUBP37 1381
1 SYS_SUBP38 1381
2 SYS_SUBP36 1531
1 SYS_SUBP37 1531
2 SYS_SUBP38 1531
1 SYS_SUBP39 1531
1 SYS_SUBP40 1531
2 SYS_SUBP33 1566
1 SYS_SUBP34 1566
1 SYS_SUBP35 1566
1 SYS_SUBP37 1566
1 SYS_SUBP38 1566
2 SYS_SUBP39 1566
3 SYS_SUBP40 1566
1 SYS_SUBP32 1567
3 SYS_SUBP33 1567
3 SYS_SUBP35 1567
3 SYS_SUBP36 1567
1 SYS_SUBP37 1567
2 SYS_SUBP38 1567
62 rows selected.
--using partition by any cluster by subpartition_key
Elapsed: 00:00:00.26
XXXX> with parts as (
2 select --+ materialize
3 data_object_id,
4 subobject_name
5 FROM
6 user_objects
7 WHERE
8 object_name = 'TEST_TAB'
9 and
10 object_type = 'TABLE SUBPARTITION'
11 )
12 SELECT
13 COUNT(*),
14 parts.subobject_name,
15 target.sid
16 FROM
17 parts,
18 test_tab tt,
19 test_tab_part_any_cluster target
20 WHERE
21 tt.tracking_id = target.tracking_id
22 and
23 parts.data_object_id = DBMS_MView.PMarker(tt.rowid)
24 GROUP BY
25 parts.subobject_name,
26 target.sid
27 ORDER BY
28 target.sid,
29 parts.subobject_name
30 /
COUNT(*) SUBOBJECT_NAME SID
11 SYS_SUBP37 1253
10 SYS_SUBP34 1268
4 SYS_SUBP31 1289
10 SYS_SUBP40 1314
7 SYS_SUBP39 1367
9 SYS_SUBP35 1377
14 SYS_SUBP36 1531
5 SYS_SUBP32 1572
13 SYS_SUBP33 1577
17 SYS_SUBP38 1609
10 rows selected.Bear in mind though that this does require a sort of the incomming dataset but does not require buffering of the output...
PLAN_TABLE_OUTPUT
Plan hash value: 2570087774
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop | TQ |IN-OUT| PQ Distrib |
| 0 | SELECT STATEMENT | | 8168 | 1722K| 24 (0)| 00:00:01 | | | | | |
| 1 | PX COORDINATOR | | | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10000 | 8168 | 1722K| 24 (0)| 00:00:01 | | | Q1,00 | P->S | QC (RAND) |
| 3 | VIEW | | 8168 | 1722K| 24 (0)| 00:00:01 | | | Q1,00 | PCWP | |
| 4 | COLLECTION ITERATOR PICKLER FETCH| TF_ANY | | | | | | | Q1,00 | PCWP | |
| 5 | SORT ORDER BY | | | | | | | | Q1,00 | PCWP | |
| 6 | PX BLOCK ITERATOR | | 100 | 4800 | 2 (0)| 00:00:01 | 1 | 10 | Q1,00 | PCWC | |
| 7 | TABLE ACCESS FULL | TEST_TAB | 100 | 4800 | 2 (0)| 00:00:01 | 1 | 20 | Q1,00 | PCWP | |
----------------------------------------------------------------------------------------------------------------------------------------------HTH
David -
How to debug "pipelined parallel enable table function" ?
Dear All,
Normally we can retrieve output from a "pipelined parallel enable table function " by using SQL statement, such as
select output from table(pipelined_function(arg1));
Can anyone alight me how to call this function purely using PL/SQL statement?
Reason for this : We have third party developer develooped a complicate "pipelined parallel enable table function" and this function is currently calling by SQL (select output from table(pipelined_function(arg1));
), I will like to debug this function, but unable to do so, so far I have done :
1) compile function with debug option
2) using Procedure Builder to debug, if I execute the above statement (select output from table(pipelined_function(arg1)); I believe, Procedure Builder will not debug a function from a SQL statement
3) try to build up a PL/SQL block, but don't know how to it.
basically I want to debug this "pipelined parallel enable table function " and don't know how to do it, any example will be greate!user2302827 wrote:
Using dbms_output is fine but too tedious. I was looking for a PL/SQL procedure builder that can help to debug a statement like this:
select output from table(pipelined_function(arg1));
Any suggestion?I don't know about any procedure builders.
There is a debugger you can use, though I'd have to think about how to invoke it to use with a pipelined function. The debugger is available in GUI tools like SQL*Developer and TOAD and can be used to walk through a program and set values -
Compare two results from the same table
i have two results from the same table that i would like to compare. below is my query and the results i want to compare
SELECT tblItemRoutingBOM.ItemRevID, tblItem.ItemID, tblItem.PartNum, tblItem.ItemName, tblItem.ManufacturerPartNum AS [Mfg Part#], tblItemRoutingBOM.Quantity
FROM tblItemRouting
INNER JOIN tblItemRoutingBOM ON tblItemRouting.ItemRoutingID = tblItemRoutingBOM.ItemRoutingID
INNER JOIN tblItem ON tblItemRoutingBOM.ItemID = tblItem.ItemID
WHERE tblItemRoutingBOM.ItemRevID in (61,70)
as you can see i am returning two records using the where clause
ItemRevID, ItemID, PartNum, ItemName, Manufacturer, Mfg Part#, Quantity
61,121,331503,.233 Aluminum Sheet,,1
70,121,331503,.233 Aluminum Sheet,,3
now what i am looking for is to combine these two together into one row with the following added. two columns for each QTY, QTY1 = 1 and QTY2 = 3 with a third column added that is the difference between the two QTY Diff = 2
Any thoughts?Here are the two statements that i want to combine, results for each are attached
SELECT tblItem.ItemID, Sum(tblItemRoutingBOM.Quantity) AS SumOfQuantity, tblItem.PartNum AS [Part #],
tblItem.ItemName, tblManufacturer.ManufacturerName AS Manufacturer, tblItem.ManufacturerPartNum AS [Mfg Part#]
FROM tblItemRouting
INNER JOIN tblItemRoutingBOM ON tblItemRouting.ItemRoutingID = tblItemRoutingBOM.ItemRoutingID
INNER JOIN tblItem ON tblItemRoutingBOM.ItemID = tblItem.ItemID
INNER JOIN tblUnits ON tblItem.UnitID = tblUnits.UnitID
LEFT JOIN tblManufacturer ON tblItem.ManufacturerID = tblManufacturer.ManufacturerID
WHERE tblItemRoutingBOM.ItemRevID=61
GROUP BY tblItem.ItemID,tblItem.PartNum,tblItem.ItemName,tblManufacturer.ManufacturerName,tblItem.ManufacturerPartNum
SELECT tblItem.ItemID, Sum(tblItemRoutingBOM.Quantity) AS Quantity, tblItem.PartNum AS [Part #],
tblItem.ItemName, tblManufacturer.ManufacturerName AS Manufacturer, tblItem.ManufacturerPartNum AS [Mfg Part#]
FROM tblItemRouting
INNER JOIN tblItemRoutingBOM ON tblItemRouting.ItemRoutingID = tblItemRoutingBOM.ItemRoutingID
INNER JOIN tblItem ON tblItemRoutingBOM.ItemID = tblItem.ItemID
INNER JOIN tblUnits ON tblItem.UnitID = tblUnits.UnitID
LEFT JOIN tblManufacturer ON tblItem.ManufacturerID = tblManufacturer.ManufacturerID
WHERE tblItemRoutingBOM.ItemRevID=70
GROUP BY tblItem.ItemID,tblItem.PartNum,tblItem.ItemName,tblManufacturer.ManufacturerName,tblItem.ManufacturerPartNum
114,11,55002,Pepsi Blue Cap,NULL,
117,5,331501,Marigold Yellow For ABS,NULL,
121,1,331503,.233 Aluminum Sheet,NULL,
125,2,331504,Velvet Vinyl .008,NULL,
114,33,55002,Pepsi Blue Cap,NULL,
117,15,331501,Marigold Yellow For ABS,NULL,
121,3,331503,.233 Aluminum Sheet,NULL,
125,6,331504,Velvet Vinyl .008,NULL,
my returned result should combine above with two extra columns (two extra columns because i have two results to combine)
114, 11, 33, 22, 55002, Pepsi Blue Cap, NULL,
117, 5, 15, 10, 331501, Marigold Yellow For ABS, NULL
121,1, 3, 2, 331503, .233 Aluminum Sheet, NULL
125, 2, 6, 4, 331504, Velvet Vinyl .008, NULL
Columns go as such, ID, QTY1 (for 61), QTY2 (for 70), Diff (QTY1-QTY2), PartNum, ItemName, Mfg, Mfg Part#
IF the results from one of those two are empty then i would see something like this
114, 11, 0, 11, 55002, Pepsi Blue Cap, NULL,
117, 5, 0, 5, 331501, Marigold Yellow For ABS, NULL
121,1, 0, 1, 331503, .233 Aluminum Sheet, NULL
125, 2, 0, 2, 331504, Velvet Vinyl .008, NULL -
How to fetch data for a struture from a cluster table
How can I fetch data for a struture, from a cluster table, based on the name of the structure?
Hi,
In order to read from Cluster DB Table use the following statement:
Syntax
IMPORT <f1> [ TO < g1 > ] <f2> [TO < g2 >] ...
FROM DATABASE <dbtab>(<ar>)
[CLIENT <cli>] ID <key>|MAJOR-ID <maid> [MINOR-ID <miid>].
This statement reads the data objects specified in the list from a cluster in the database <dbtab>.
You must declare <dbtab> using a TABLES statement. If you do not use the TO <gi> option, the
data object <fi> in the database is assigned to the data object in the program with the same
name. If you do use the option, the data object <fi> is read from the database into the field <gi>.
For <ar>, enter the two-character area ID for the cluster in the database. The name <key>
identifies the data in the database. Its maximum length depends on the length of the name field
in <dbtab>.
The CLIENT <cli> option allows you to disable the automatic client handling of a client-specific cluster database, and specify the client yourself. The addition must always come directly after the name of the database.
For Eg:
PROGRAM SAPMZTS3.
TABLES INDX.
DATA: BEGIN OF JTAB OCCURS 100,
COL1 TYPE I,
COL2 TYPE I,
END OF JTAB.
IMPORT ITAB TO JTAB FROM DATABASE INDX(HK) ID 'Table'.
WRITE: / 'AEDAT:', INDX-AEDAT,
/ 'USERA:', INDX-USERA,
/ 'PGMID:', INDX-PGMID.
SKIP.
WRITE 'JTAB:'.
LOOP AT JTAB FROM 1 TO 5.
WRITE: / JTAB-COL1, JTAB-COL2.
ENDLOOP.
Regards,
Neha
Edited by: Neha Shukla on Mar 12, 2009 1:35 PM -
Query regarding the data type for fetcing records from multiple ODS tables
hey guys;
i have a query regarding the data type for fetcing records from multiple ODS tables.
if i have 2 table with a same column name then in the datatype under parent row node i cant add 2 nodes with the same name.
can any one help with some suggestion.Hi Mudit,
One option would be to go as mentioned by Padamja , prefxing the table name to the column name or another would be to use the AS keyoword in your SQL statement.
AS is used to rename the column name when data is being selected from your DB.
So, the query Select ename as empname from emptable will return the data with column name as empname.
Regards,
Bhavesh -
Performance issues with pipelined table functions
I am testing pipelined table functions to be able to re-use the <font face="courier">base_query</font> function. Contrary to my understanding, the <font face="courier">with_pipeline</font> procedure runs 6 time slower than the legacy <font face="courier">no_pipeline</font> procedure. Am I missing something? The <font face="courier">processor</font> function is from [url http://www.oracle-developer.net/display.php?id=429]improving performance with pipelined table functions .
Edit: The underlying query returns 500,000 rows in about 3 minutes. So there are are no performance issues with the query itself.
Many thanks in advance.
CREATE OR REPLACE PACKAGE pipeline_example
IS
TYPE resultset_typ IS REF CURSOR;
TYPE row_typ IS RECORD (colC VARCHAR2(200), colD VARCHAR2(200), colE VARCHAR2(200));
TYPE table_typ IS TABLE OF row_typ;
FUNCTION base_query (argA IN VARCHAR2, argB IN VARCHAR2)
RETURN resultset_typ;
c_default_limit CONSTANT PLS_INTEGER := 100;
FUNCTION processor (
p_source_data IN resultset_typ,
p_limit_size IN PLS_INTEGER DEFAULT c_default_limit)
RETURN table_typ
PIPELINED
PARALLEL_ENABLE(PARTITION p_source_data BY ANY);
PROCEDURE with_pipeline (argA IN VARCHAR2,
argB IN VARCHAR2,
o_resultset OUT resultset_typ);
PROCEDURE no_pipeline (argA IN VARCHAR2,
argB IN VARCHAR2,
o_resultset OUT resultset_typ);
END pipeline_example;
CREATE OR REPLACE PACKAGE BODY pipeline_example
IS
FUNCTION base_query (argA IN VARCHAR2, argB IN VARCHAR2)
RETURN resultset_typ
IS
o_resultset resultset_typ;
BEGIN
OPEN o_resultset FOR
SELECT colC, colD, colE
FROM some_table
WHERE colA = ArgA AND colB = argB;
RETURN o_resultset;
END base_query;
FUNCTION processor (
p_source_data IN resultset_typ,
p_limit_size IN PLS_INTEGER DEFAULT c_default_limit)
RETURN table_typ
PIPELINED
PARALLEL_ENABLE(PARTITION p_source_data BY ANY)
IS
aa_source_data table_typ;-- := table_typ ();
BEGIN
LOOP
FETCH p_source_data
BULK COLLECT INTO aa_source_data
LIMIT p_limit_size;
EXIT WHEN aa_source_data.COUNT = 0;
/* Process the batch of (p_limit_size) records... */
FOR i IN 1 .. aa_source_data.COUNT
LOOP
PIPE ROW (aa_source_data (i));
END LOOP;
END LOOP;
CLOSE p_source_data;
RETURN;
END processor;
PROCEDURE with_pipeline (argA IN VARCHAR2,
argB IN VARCHAR2,
o_resultset OUT resultset_typ)
IS
BEGIN
OPEN o_resultset FOR
SELECT /*+ PARALLEL(t, 5) */ colC,
SUM (CASE WHEN colD > colE AND colE != '0' THEN colD / ColE END)de,
SUM (CASE WHEN colE > colD AND colD != '0' THEN colE / ColD END)ed,
SUM (CASE WHEN colD = colE AND colD != '0' THEN '1' END) de_one,
SUM (CASE WHEN colD = '0' OR colE = '0' THEN '0' END) de_zero
FROM TABLE (processor (base_query (argA, argB),100)) t
GROUP BY colC
ORDER BY colC
END with_pipeline;
PROCEDURE no_pipeline (argA IN VARCHAR2,
argB IN VARCHAR2,
o_resultset OUT resultset_typ)
IS
BEGIN
OPEN o_resultset FOR
SELECT colC,
SUM (CASE WHEN colD > colE AND colE != '0' THEN colD / ColE END)de,
SUM (CASE WHEN colE > colD AND colD != '0' THEN colE / ColD END)ed,
SUM (CASE WHEN colD = colE AND colD != '0' THEN 1 END) de_one,
SUM (CASE WHEN colD = '0' OR colE = '0' THEN '0' END) de_zero
FROM (SELECT colC, colD, colE
FROM some_table
WHERE colA = ArgA AND colB = argB)
GROUP BY colC
ORDER BY colC;
END no_pipeline;
END pipeline_example;
ALTER PACKAGE pipeline_example COMPILE;Edited by: Earthlink on Nov 14, 2010 9:47 AM
Edited by: Earthlink on Nov 14, 2010 11:31 AM
Edited by: Earthlink on Nov 14, 2010 11:32 AM
Edited by: Earthlink on Nov 20, 2010 12:04 PM
Edited by: Earthlink on Nov 20, 2010 12:54 PMEarthlink wrote:
Contrary to my understanding, the <font face="courier">with_pipeline</font> procedure runs 6 time slower than the legacy <font face="courier">no_pipeline</font> procedure. Am I missing something? Well, we're missing a lot here.
Like:
- a database version
- how did you test
- what data do you have, how is it distributed, indexed
and so on.
If you want to find out what's going on then use a TRACE with wait events.
All nessecary steps are explained in these threads:
HOW TO: Post a SQL statement tuning request - template posting
http://oracle-randolf.blogspot.com/2009/02/basic-sql-statement-performance.html
Another nice one is RUNSTATS:
http://asktom.oracle.com/pls/asktom/ASKTOM.download_file?p_file=6551378329289980701 -
Interactive report on view based on pipelined table function.
Hi,
I want to build an Interactive Report on a view.
The view definition contains a select on a pipelined table function. I use context functionality to pass paramaters to the pipelined table function.
A plain select * from #my_view# in SqlPlus results in 121 different rows.
However, If I base my Interactive report on this view, I get 15 repeated rows (all the same).
Is it possible to use pipelined table functionality on an Interactive report? I can't seem to get it working.
If I use the following approach (http://rakeshjsr.blogspot.nl/2010/10/oracle-apex-interactive-report-based-on.html) I do get results, but I can't use this solution for a reason that's not relevant.Hello,
Is it possible to use pipelined table functionality on an Interactive report? I can't seem to get it working. I have used it in one instance and it works fine. However I was passing the values to pipe-lined function directly.
IR Query..
SELECT * FROM TABLE(fn_pipeline(:P1_ITEM_NAME))Call pipe-lined function from IR query directly (instead of using view)
Try sending values to Pipe-lined function directly. In-case if the problem is with setting and getting values from the context?
Regards,
Hari -
Query performance improvement using pipelined table function
Hi,
I have got two select queries one is like...
select * from table
another is using pielined table function
select *
from table(pipelined_function(cursor(select * from table)))
which query will return result set more faster????????
suggest methods for retrieving dataset more faster (using pipelined table function) than a normal select query.
rgds
somyCompare the performance between these solutions:
create table big as select * from all_objects;
First test the performance of a normal select statement:
begin
for r in (select * from big) loop
null;
end loop;
end;
/Second a pipelined function:
create type rc_vars as object
(OWNER VARCHAR2(30)
,OBJECT_NAME VARCHAR2(30));
create or replace type rc_vars_table as table of rc_vars ;
create or replace
function rc_get_vars
return rc_vars_table
pipelined
as
cursor c_aobj
is
select owner, object_name
from big;
l_aobj c_aobj%rowtype;
begin
for r_aobj in c_aobj loop
pipe row(rc_vars(r_aobj.owner,r_aobj.object_name));
end loop;
return;
end;
/Test the performance of the pipelined function:
begin
for r in (select * from table(rc_get_vars)) loop
null;
end loop;
end;
/On my system the simple select-statement is 20 times faster.
Correction: It is 10 times faster, not 20.
Message was edited by:
wateenmooiedag -
Using Pipeline Table functions with other tables
I am on DB 11.2.0.2 and have sparingly used pipelined table functions but am considering it for a project that has some fairly big (lots of rows) sized tables. In my tests, selecting from just the pipelined table perform pretty well (whether it is directly from the pipleined table or the view I created on top of it). Where I start to see some degregation when I try to join the pipelined tabe view to other tables and add where conditions.
ie:
SELECT A.empno, A.empname, A.job, B.sal
FROM EMP_VIEW A, EMP B
WHERE A.empno = B.empno AND
B.mgr = '7839'
I have seen some articles and blogs that mention this as a cardinality issue, and offer some undocumented methods to try and combat.
Can someone please give me some advice or tips on this. Thanks!
I have created a simple example using the emp table below to help illustrate what I am doing.
DROP TYPE EMP_TYPE;
DROP TYPE EMP_SEQ;
CREATE OR REPLACE TYPE EMP_SEQ AS OBJECT
( EMPNO NUMBER(10),
ENAME VARCHAR2(100),
JOB VARCHAR2(100));
CREATE OR REPLACE TYPE EMP_TYPE AS TABLE OF EMP_SEQ;
CREATE OR REPLACE FUNCTION get_emp return EMP_TYPE PIPELINED AS
BEGIN
FOR cur IN (SELECT
empno,
ename,
job
FROM emp
LOOP
PIPE ROW(EMP_SEQ(cur.empno,
cur.ename,
cur.job));
END LOOP;
RETURN;
END get_emp;
create OR REPLACE view EMP_VIEW as select * from table(get_emp());
SELECT A.empno, A.empname, A.job, B.sal
FROM EMP_VIEW A, EMP B
WHERE A.empno = B.empno AND
B.mgr = '7839'I am on DB 11.2.0.2 and have sparingly used pipelined table functions but am considering it for a project that has some fairly big (lots of rows) sized tables
Which begs the question: WHY? What PROBLEM are you trying to solve and what makes you think using pipelined table functions is the best way to solve that problem?
The lack of information about cardinality is the likely root of the degradation you noticed as already mentioned.
But that should be a red flag about pipelined functions in general. PIPELINED functions hide virtually ALL KNOWLEDGE about the result set that is produced; cardinality is just the tip of the iceberg. Those functions pretty much say 'here is a result set' without ANY information about the number of rows (cardinality), distinct values for any columns, nullability of any columns, constraints that might apply to any columns (foreign key, primary key) and so on.
If you are going to hide all of that information from Oracle that would normally be used to help optimize queries and select the appropriate execution plan you need to have a VERY good reason.
The use of PIPELINED functions should be reserved for those use cases where ordinary SQL and PL/SQL cannot get the job done. That is they are a 'special case' solution.
The classic use case for those functions is for the transform stage of ETL where multiple pipelined functions are chained together: one function feeds its rows to the next function which feeds its rows to another and so on. Each of those 'chained' functions is roughly analogous to a full table scan of the data that often does not need to be joined to other data except perhaps low volumn lookup tables where the data may even be cached.
I suggest that any exploratory or prototyping work you do use standard relational tables until such point as you run into a problem whose solution might require PIPELINED functions to solve. -
Pipeline Table Function returning a fraction of data
My current project involves migrating an Oracle database to a new structure based on the new client application requirements. I would like to use pipelined table functions as it seems as though that would provide the best performance.
The first table has about 65 fields, about 75% of which require some type of recoding for the new app. I have written a function for each transformation and have all of these functions stored in a package. If I do:
create new_table as select (
pkg_name.function1(old_field1),
pkg_name.function2(old_field2),
pkg_name.function3(old_field3),
it runs with out any errors but takes about 3 1/2 hours. There are a little more than 10 million rows in the table.
I wrote a function that is passed the old table as a cursor, runs all the functions for the transformations and then pipes the new row back to the insert statement that called the function. It is incredibly fast but only returns .025% of the data (about 50 rows out of my sample table of 200,000). It does not throw any errors.
So I am trying to determine what is going on. Perhaps one of my functions has a bug. If there was would cause the row to be kicked out? There are 40 or so functions so tracking this down has been a bit of a bear.
Any advice as to how I might resolve this would be much appreciated.
Thanks
Dan. I would like to use pipelined table functions as it seems as though that would provide the best performanceUh huh...
it runs with out any errors but takes about 3 1/2 hours. There are a little more than 10 million rows in the table.Not the first time a lovely theory has been killed by an ugly fact. Did you do any bench marks to see whether the pipelined functions did offer performance benefits over doing it some other way?
From the context of your comments I think you are trying to a populate a new table from a single old table. Is this the case? If so I would have thought a straightforward CTAS with normal functions would be more appropriate: pipelined functions are really meant for situations in which one input produced more than one output. Anyway, ifr we are to help you I think you need to give us more details about how this process works and post a sample transformation function.
There are 40 or so functions so tracking this down has been a bit of a bear.The teaching is: we should code one function and get that working before moving on to the next one. Which might not seem like a helpful thing to say, but the best lesson is often "I'll do it differently next time".
Cheers, APC -
Distributed queries+pipelined table function
HI friends,
can i get better performance for distributed queries if i use pipelined table function.I have got my data distribued across three different databases.
thanx
somyYou will need to grant EXECUTE access on the pipelined table function to whatever users want it. When other users call this function, they may need to prefix the schema owner (i.e. <<owner>>.getValue('001') ) unless you've set up the appropriate synonym.
What version of SQL*Plus do you have on the NT machine?
Justin
Distributed Database Consulting, Inc.
http://www.ddbcinc.com/askDDBC -
Best practices for logging results from Looped steps
Hi all
I would like to start a discussion to document best practices for logging results (to reports and databases) from Looped Steps
As an application example - let's say you are developing a test for one of NI's analog input or output cards and need to measure a voltage across multiple inputs or outputs.
One way to do that would be to create a sequence that switches the appropriate signals and performs a "Voltage Measurement" test in a loop.
What are your techniques for keeping track of the individual measurements so that they can be traced to the individual signal paths that are being measured?
I have used a variety of techniques such as
i )creating a custom step type that generates unique identifiers for each iteration of the loop. This required some customization to the results processing . Also the sequence developer had to include code to ensure that a unique identifier was generated for each iteration
ii) Adding an input parameter to the test function/vi, passing loop iteration to it and adding this to Additional results parameters to log.I have attached a simple example (LV 2012 and TS 2012) that includes steps inside a loop structure as well as a looped test.
If you enable both database and report generation, you will see the following:
1) The numeric limit test in the for loop always generates the same name in the report and database which makes it difficult to determine the result of a particular iteration
2) The Max voltage test report includes the paramater as an additional result but the database does not include any differentiating information
3) The Looped Limit test generates both uniques reports and database entries - you can easily see what the result for each iteration is.
As mentioned, I am seeking to start a discussion for how others handle results for steps inside loops. The only way I have been able to accomplish a result similar to that of the Looped step (unique results and database entry for each iteration of the loop) is to modify the process model results processing.
Attachments:
test.vi 27 KB
Sequence File 2.seq 9 KB
Maybe you are looking for
-
Service order creation condition
Hi, Could anyone please tell me how can we insert a condition so that one type of service order can not be created after creation of service order of another type. Suppose if A and B type of service orders(sub-order of C) are to created sequentially.
-
Time machine backup seems to only have a Sparsebundle folder
I have had my MacBook for a couple of years and have only just got round to backing it up. I set off Time Machine lat night for a 200Gb backup. This morning I checked it and the Time machine backup seems to only have a Sparsebundle folder of 1000s of
-
Hey i hav an hp pavilion g series n i would like to kno how use the function keys without having to
hey i hav an hp pavilion g series n i would like to kno how use the function keys without having to click on the fn + function key
-
Import Utility Without Duplication
I planning to export one Schema from an Oracle database (let say DB1) and I want to import it into another Oracle database (let say DB2). However, DB2 already has the exported Schema along with its contents (Tables and its data/records), but the data
-
Help about location based classes
Helo. i need some help about location classes(i use J2ME). i want to develop a gps compass application. when i connected to gps, orientation is shown on a compass.(for example if i go to northeast direction, compass indicator shows the direction betw