DB links

A recent thread on the ODC SQL & PL/SQL forum raised the question of how data moves across a database link – is it row by row or array based (or other)? It’s a good question as it’s fairly common knowledge that distributed queries can be very much slower than the same query when executed on an equivalent set of local tables, so it’s worth having a little poke around to see if there’s anything in the way the data moves across the link that could be tweaked to improve performance.

It’s fairly easy to get some ideas by running a couple of queries and checking session activity stats and wait events – so here’s a little test conducted between a database running 12.2.0.1 and a database running 11.2.0.4. For this type of testing it’s probably sensible to use two database rather than faking things with a loopback database link in case the loopback introduces some special features (a.k.a. dirty tricks).

rem
rem     Script:         db_link_usage.sql
rem     Author:         Jonathan Lewis
rem     Dated:          July 2019
rem
rem     Last tested 
rem             12.2.0.1 -> 11.2.0.4
rem

prompt  ==================================
prompt  SQL*Plus 20,000 rows, arraysize 20
prompt  ==================================

set feedback only
set arraysize 20

execute snap_my_stats.start_snap
execute snap_events.start_snap
select * from all_objects@d11g where rownum <= 20000;

set feedback on
execute snap_events.end_snap
execute snap_my_stats.end_snap

prompt  ====================================
prompt  SQL*Plus 20,000 rows, arraysize 2000
prompt  ====================================

set feedback only
set arraysize 2000

execute snap_my_stats.start_snap
execute snap_events.start_snap
select * from all_objects@d11g where rownum <= 20000;

set feedback on
execute snap_events.end_snap
execute snap_my_stats.end_snap

All I’ve done is select 20,000 rows from view all_objects in a remote database with two different settings for the array fetch size. The “feedback only” option is a 12c option that means the session doesn’t output the data, it shows only the final “N rows selected” report. The two “snap” packages are some code I wrote a couple of decades ago to report changes in my session’s activity stats and wait events.

It turned out that there was very little of interest in the session activity stats although the impact of SQL*Net compression is always worth a quick reminder (here and here) on the other hand the wait events (specifically the wait counts) are most revealing:

================================== 
SQL*Plus 20,000 rows, arraysize 20
================================== 
Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                       1,004           0          15.69        .016     414,828
SQL*Net message to dblink                         1,004           0           0.04        .000           0
SQL*Net message from dblink                       1,004           0          61.02        .061          11
SQL*Net message to client                         1,004           0           0.11        .000           0

====================================
SQL*Plus 20,000 rows, arraysize 2000
====================================
Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                          14           0           4.88        .349     414,828
SQL*Net message to dblink                            13           0           0.00        .000           0
SQL*Net message from dblink                          13           0           7.72        .594          11
SQL*Net message to client                            14           0           0.00        .000           0

SQL*Net more data from dblink                        96           0           6.26        .065           6
SQL*Net more data to client                          96           0           0.16        .002           4

Doing a little arithmetic (and allowing for a few “metadata” messages that introduce small variations in the numbers), we can see that when we fetched the 20,000 rows with an arraysize of 20 this turned into 1,000 (fetch) calls from SQL*Plus to the server, with a matching 1,000 calls from the local server to the remote server. When the arraysize goes up to 2,000 rows, though, the SDU (session data unit) setting for SQL*Net is too small to hold the whole of a single fetch and we see a single fetch from SQL*Plus turning into one “SQL*Net message to client” accompanied by 7 or 8 “SQ:(Net more data to client” with exactly the same pattern of conversation between the local server and the remote server. You could imagine the conversation as something like:

• Client to local server: “give me 2,000 rows”
• local server to remote server: “give me 2,000 rows”
• remote server to local server: “Here, I can manage 120 rows as the first installment”
• local server to client: “Here’s a first installment”
• Client to local server: “Got it, where’s the rest?”     *** but no message recorded in the session activity stats
• Local server to remote server: “Where’s the rest?”
• Remote server to local server: “You want more – here’s another 120 rows”
• Local server to client: “Here’s a second installment”
• Client to local server: “Got it, where’s the rest?”     *** but no message recorded in the session activity stats
• etc.

In this example the volume of data sent back to the client in each message was limited by the SDU size negotiated between the local server and the remote server as the link opens. In my case I happened to have the default (apparently 8KB) as the SDU_SIZE in the 12c and a non-default 32KB setting in the 11g sqlnet.ora, so the negotiated SDU between servers was 8KB (the lower of the two).

Here’s what the figures looked like after I had restarted with the SDU set to 16KB in the 12c tnsnames.ora, and 16KB in the 11g listener.ora:

==================================
SQL*Plus 20,000 rows, arraysize 20
==================================
Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                       1,004           0          24.23        .024         588
SQL*Net message to dblink                         1,009           0           0.06        .000           0
SQL*Net message from dblink                       1,010           0          77.76        .077           4
SQL*Net message to client                         1,004           0           0.15        .000           0

====================================
SQL*Plus 20,000 rows, arraysize 2000
====================================
Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                          14           0           1.61        .115         588
SQL*Net message to dblink                            13           0           0.00        .000           0
SQL*Net message from dblink                          13           0           4.21        .324           4
SQL*Net message to client                            14           0           0.00        .000           0

SQL*Net more data from dblink                        45           0          13.53        .301           6
SQL*Net more data to client                          45           0           0.13        .003           0

The first set of figures (arraysize 20) don’t change. If an 8KB SDU is large enough to hold an array of 20 rows then it’s not going to make a difference when the SDU is increased. In the second set of figures we see that for each “SQL*Net message from dblink” we now have roughly 3 “SQL*Net more data from dblink” (with matching counts for “SQL*Net more data to client”). With an SDU of double the size it now takes a total of roughly 4 packets to transmit the array fetch rather than the 8 or 9 we saw with the smaller SDU size.

As a final test with SQL*Plus, what happens if we set the SDU size at 32K for the 12c database (and that means for the client calling in and the server calling out) and 8KB for the 11g database? The client negotiates a 32KB SDU with the 12c server, but the 12c server negotiates an 8KB SDU with the 11g server. Here are the stats for just the test with the large fetch size in this case:

====================================
SQL*Plus 20,000 rows, arraysize 2000
====================================
Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                          14           0           5.30        .379     214,570
SQL*Net message to dblink                            13           0           0.00        .000           0
SQL*Net message from dblink                          13           0           4.09        .314          13
SQL*Net message to client                            14           0           0.00        .000           0

SQL*Net more data from dblink                        96           0          14.46        .151           6
SQL*Net more data to client                          20           0           0.08        .004           0

We get the same 10 (plus a few) “message to/from client/dblink”, but now the “more data” waits are dramatically different. When the client calls for the “next fetch” the local server has to call the remote server 4 or 5 times to get enough 8KB data packets to fill a single 32KB packet to return to the client. You can confirm this (and some of my descriptions of the earlier behaviour) by enabling extended tracing for wait states. Here’s an extract of 5 consecutive lines from a suitable trace file (you have to ignore the “#bytes” parameter on the “SQL*Net more data from dblink” waits, they’ve always been garbage:

WAIT #140028098478296: nam='SQL*Net more data from dblink' ela= 2793 driver id=1413697536 #bytes=7 p3=0 obj#=-1 tim=703316387674
WAIT #140028098478296: nam='SQL*Net more data from dblink' ela= 34 driver id=1413697536 #bytes=1 p3=0 obj#=-1 tim=703316388447
WAIT #140028098478296: nam='SQL*Net more data from dblink' ela= 8 driver id=1413697536 #bytes=2 p3=0 obj#=-1 tim=703316389134
WAIT #140028098478296: nam='SQL*Net more data from dblink' ela= 16 driver id=1413697536 #bytes=12 p3=0 obj#=-1 tim=703316389818
WAIT #140028098478296: nam='SQL*Net more data to client' ela= 73 driver id=1413697536 #bytes=32671 p3=0 obj#=-1 tim=703316389960

Summary Note

Oracle will use the negotiated SDU to do array fetches across a database link. Since a larger arraysize can (for large volumes of data) reduce the total work load on the remote database and on the network you may want the local server to have a specifically crafted tnsnames.ora entry and the remote server to expose a specific service with matching SDU size to help improve the efficiency of transferring data between two databases.

Lagniappe

If you want to look at other cases of how array fetching and SDU sizes interact, here are a couple of examples of using PL/SQL to execute SQL that fetches data across database links. The first is a simple, embedded “cursor for loop” that (apparently) does “row by row” procssing – although an enhancement appeared many versions ago to make this type of code use array fetching of 100 rows under the covers. The second demonstrates the effects of an explicit cursor with “fetch, bulk collect, limit”:

execute snap_events.start_snap

declare
        m_total number := 0;
        m_ct    number := 0;
begin
        for c1 in (select * from all_objects@d11g where rownum < 20000) loop
                m_total := m_total + c1.object_id;
                m_ct    := m_ct + 1;
        end loop;
end;
/

execute snap_events.end_snap

execute snap_events.start_snap

declare

        cursor c1 is  select * from all_objects@d11g where rownum <= 20000;

        type c1_array is table of c1%rowtype index by binary_integer;
        m_tab c1_array;

        m_ct number := 0;

begin
        open c1;
        loop
                fetch   c1
                bulk collect
                into    m_tab
                limit   1000
                ;

                exit when c1%notfound;
                m_ct := m_ct + 1;

        end loop;
        close c1;
end;
/

execute snap_events.end_snap

Checking the session events for these two test we see the following when the SDU_SIZE has been set at 16KB at both ends of the database link (which means the negotiated SDU will be 16KB):

Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                           4           0           0.66        .165         588 
SQL*Net message to dblink                           203           0           0.05        .000           0
SQL*Net message from dblink                         203           0          38.51        .190           5
SQL*Net message to client                             4           0           0.00        .000           0

Event                                             Waits   Time_outs           Csec    Avg Csec    Max Csec
-----                                             -----   ---------           ----    --------    --------
SQL*Net message from client                           4           0           0.08        .021         588
SQL*Net message to dblink                            24           0           0.01        .000           0
SQL*Net message from dblink                          24           0           9.19        .383           5
SQL*Net more data from dblink                        40           0           8.47        .212           6
SQL*Net message to client                             4           0           0.00        .000           0

The critical information we can see in the first example is that getting 20,000 rows from the remote database “row by row” takes 200 (plus noise) round-trips – showing that Oracle really is converting our single-row processing loop in array fetches of 100 rows.

The second example shows us that (allowing for a little noise, again) it takes 20 messages to the remote database to fetch 20,000 rows – so 1,000 rows per message – but the SDU size is too small and we have to send each array in three consecutive packets, the “message from dblink” wait and two “more data from dblink” waits.

Footnote

I did point out that the session activity stats for theses tests weren’t interesting. There is, however, one detail worth mentioning since you could otherwise be fooled into thinking the number of packet travelling between the databases was smaller than it really was. There are two statistics about network roundtrips:

Name                                         Value
----                                         -----
SQL*Net roundtrips to/from client               18         
SQL*Net roundtrips to/from dblink               13

A “roundtrip” is a “message to / message from” pair. It doesn’t account for traffic due to “more data” being requested or sent.

 

20 Indexes

If your system had to do a lot of distributed queries there’s a limit on indexes that might affect performance: when deriving an execution plan for a distributed query the optimizer will consider a maximum of twenty indexes on each remote table. if you have any tables with a ridiculous number of indexes (various 3rd party accounting and CRM systems spring to mind) and if you drop and recreate indexes on those tables in the wrong order then execution plans may change for the simple reason that the optimizer is considering a different subset of the available indexes.

Although the limit is stated in the manuals (a few lines into a section on managing statement transparency) there is no indication about which 20 indexes the optimizer is likely to choose – a couple of experiments, with tracing enabled and shared pool flushes, gives a fairly strong indication that it’s the last 20 indexes created (or, to be more explicit, the ones with the 20 highest object_id values).

Here’s a little code to help demonstrate the point – first just the table and index creation

rem
rem	Script:		indexes_20.sql
rem	Author:		Jonathan Lewis
rem	Dated:		Apr 2008
rem
rem	Last tested 
rem		12.2.0.1
rem

create table t1
as
with generator as (
	select	--+ materialize
		rownum 	id
	from	all_objects 
	where	rownum <= 3000 -- > comment to avoid WordPress format issue
)
select
	mod(rownum,trunc(5000/1))	n01,
	mod(rownum,trunc(5000/2))	n02,
	mod(rownum,trunc(5000/3))	n03,
	mod(rownum,trunc(5000/4))	n04,
	mod(rownum,trunc(5000/5))	n05,
	mod(rownum,trunc(5000/6))	n06,
	mod(rownum,trunc(5000/7))	n07,
	mod(rownum,trunc(5000/8))	n08,
	mod(rownum,trunc(5000/9))	n09,
	mod(rownum,trunc(5000/10))	n10,
	mod(rownum,trunc(5000/11))	n11,
	mod(rownum,trunc(5000/12))	n12,
	mod(rownum,trunc(5000/13))	n13,
	mod(rownum,trunc(5000/14))	n14,
	mod(rownum,trunc(5000/15))	n15,
	mod(rownum,trunc(5000/16))	n16,
	mod(rownum,trunc(5000/17))	n17,
	mod(rownum,trunc(5000/18))	n18,
	mod(rownum,trunc(5000/19))	n19,
	mod(rownum,trunc(5000/20))	n20,
	mod(rownum,trunc(5000/21))	n21,
	mod(rownum,trunc(5000/22))	n22,
	mod(rownum,trunc(5000/23))	n23,
	mod(rownum,trunc(5000/24))	n24,
	rownum				id,
	rpad('x',40)			padding
from
	generator	v1,
	generator	v2
where
	rownum <= 1e5 -- > comment to avoid WordPress format issue
;

--
-- Typo, I missed the semi-colon at the end of this line.
-- See comment 3.
--

alter table t1 add constraint t1_pk primary key(id)

create table t2
as
with generator as (
	select	--+ materialize
		rownum 	id
	from	all_objects 
	where	rownum <= 3000 -- > comment to avoid WordPress format issue
)
select
	mod(rownum,trunc(5000/1))	n01,
	mod(rownum,trunc(5000/2))	n02,
	mod(rownum,trunc(5000/3))	n03,
	mod(rownum,trunc(5000/4))	n04,
	mod(rownum,trunc(5000/5))	n05,
	mod(rownum,trunc(5000/6))	n06,
	mod(rownum,trunc(5000/7))	n07,
	mod(rownum,trunc(5000/8))	n08,
	mod(rownum,trunc(5000/9))	n09,
	mod(rownum,trunc(5000/10))	n10,
	mod(rownum,trunc(5000/11))	n11,
	mod(rownum,trunc(5000/12))	n12,
	mod(rownum,trunc(5000/13))	n13,
	mod(rownum,trunc(5000/14))	n14,
	mod(rownum,trunc(5000/15))	n15,
	mod(rownum,trunc(5000/16))	n16,
	mod(rownum,trunc(5000/17))	n17,
	mod(rownum,trunc(5000/18))	n18,
	mod(rownum,trunc(5000/19))	n19,
	mod(rownum,trunc(5000/20))	n20,
	mod(rownum,trunc(5000/21))	n21,
	mod(rownum,trunc(5000/22))	n22,
	mod(rownum,trunc(5000/23))	n23,
	mod(rownum,trunc(5000/24))	n24,
	rownum				id,
	rpad('x',40)			padding
from
	generator	v1,
	generator	v2
where
	rownum <= 1e5 -- > comment to avoid WordPress format issue
;

create index t2_a21 on t2(n21);
create index t2_a22 on t2(n22);
create index t2_a23 on t2(n23);
create index t2_a24 on t2(n24);

create index t2_z01 on t2(n01);
create index t2_z02 on t2(n02);
create index t2_z03 on t2(n03);
create index t2_z04 on t2(n04);
create index t2_z05 on t2(n05);
create index t2_z06 on t2(n06);
create index t2_z07 on t2(n07);
create index t2_z08 on t2(n08);
create index t2_z09 on t2(n09);
create index t2_z10 on t2(n10);

create index t2_i11 on t2(n11);
create index t2_i12 on t2(n12);
create index t2_i13 on t2(n13);
create index t2_i14 on t2(n14);
create index t2_i15 on t2(n15);
create index t2_i16 on t2(n16);
create index t2_i17 on t2(n17);
create index t2_i18 on t2(n18);
create index t2_i19 on t2(n19);
create index t2_i20 on t2(n20);

alter index t2_a21 rebuild;
alter index t2_a22 rebuild;
alter index t2_a23 rebuild;
alter index t2_a24 rebuild;
 

begin
        dbms_stats.gather_table_stats(
                ownname 	 => user,
		tabname		 =>'t1',
		method_opt 	 => 'for all columns size 1',
		cascade		 => true
	);

	dbms_stats.gather_table_stats(
		ownname		 => user,
		tabname		 =>'t2',
		method_opt 	 => 'for all columns size 1',
		cascade		 => true
	);

end;
/

I’m going to use a loopback database link to join “local” table t1 to “remote” table t2 on all 24 of the nXX columns. I’ve created indexes on all the columns, messing around with index names, order of creation, and rebuilding, to cover possible selection criteria such as alphabetical order, ordering by data_object_id (rather than object_id), even ordering by name of indexed columns(!).

Now the code to run a test:

define m_target=orcl@loopback

alter session set events '10053 trace name context forever';
set serveroutput off

select
	t1.id,
	t2.id,
	t2.padding
from
	t1			t1,
	t2@&m_target		t2
where
	t1.id = 99
and	t2.n01 = t1.n01
and	t2.n02 = t1.n02
and	t2.n03 = t1.n03
and	t2.n04 = t1.n04
and	t2.n05 = t1.n05
and	t2.n06 = t1.n06
and	t2.n07 = t1.n07
and	t2.n08 = t1.n08
and	t2.n09 = t1.n09
and	t2.n10 = t1.n10
/*			*/
and	t2.n11 = t1.n11
and	t2.n12 = t1.n12
and	t2.n13 = t1.n13
and	t2.n14 = t1.n14
and	t2.n15 = t1.n15
and	t2.n16 = t1.n16
and	t2.n17 = t1.n17
and	t2.n18 = t1.n18
and	t2.n19 = t1.n19
and	t2.n20 = t1.n20
/*			*/
and	t2.n21 = t1.n21
and	t2.n22 = t1.n22
and	t2.n23 = t1.n23
and	t2.n24 = t1.n24
;

select * from table(dbms_xplan.display_cursor(null,null,'outline'));

alter session set events '10053 trace name context off';

I’ve used a substitution variable for the name of the database link – it’s a convenience I have with all my distributed tests, a list of possible defines at the top of the script depending on which database I happen to be using at the time – then enabled the optimizer (10053) trace, set serveroutput off so that I can pull the execution plan from memory most easily, then executed the query.

Here’s the execution plan – including the Remote section and Outline.

-------------------------------------------------------------------------------------------
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     | Inst   |IN-OUT|
-------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |      |       |       |   270 (100)|          |        |      |
|   1 |  NESTED LOOPS      |      |     1 |   243 |   270   (6)| 00:00:01 |        |      |
|*  2 |   TABLE ACCESS FULL| T1   |     1 |   101 |   268   (6)| 00:00:01 |        |      |
|   3 |   REMOTE           | T2   |     1 |   142 |     2   (0)| 00:00:01 | ORCL@~ | R->S |
-------------------------------------------------------------------------------------------


Outline Data
-------------
  /*+
      BEGIN_OUTLINE_DATA
      IGNORE_OPTIM_EMBEDDED_HINTS
      OPTIMIZER_FEATURES_ENABLE('12.2.0.1')
      DB_VERSION('12.2.0.1')
      ALL_ROWS
      OUTLINE_LEAF(@"SEL$1")
      FULL(@"SEL$1" "T1"@"SEL$1")
      FULL(@"SEL$1" "T2"@"SEL$1")
      LEADING(@"SEL$1" "T1"@"SEL$1" "T2"@"SEL$1")
      USE_NL(@"SEL$1" "T2"@"SEL$1")
      END_OUTLINE_DATA
  */

Predicate Information (identified by operation id):
---------------------------------------------------
   2 - filter("T1"."ID"=99)

Remote SQL Information (identified by operation id):
----------------------------------------------------
   3 - SELECT "N01","N02","N03","N04","N05","N06","N07","N08","N09","N10","N11","N1
       2","N13","N14","N15","N16","N17","N18","N19","N20","N21","N22","N23","N24","ID","PA
       DDING" FROM "T2" "T2" WHERE "N01"=:1 AND "N02"=:2 AND "N03"=:3 AND "N04"=:4 AND
       "N05"=:5 AND "N06"=:6 AND "N07"=:7 AND "N08"=:8 AND "N09"=:9 AND "N10"=:10 AND
       "N11"=:11 AND "N12"=:12 AND "N13"=:13 AND "N14"=:14 AND "N15"=:15 AND "N16"=:16
       AND "N17"=:17 AND "N18"=:18 AND "N19"=:19 AND "N20"=:20 AND "N21"=:21 AND
       "N22"=:22 AND "N23"=:23 AND "N24"=:24 (accessing 'ORCL@LOOPBACK' )

There’s a little oddity with the plan – specifically in the Outline: there’s a “full(t2)” hint which is clearly inappropriate and isn’t consistent with the cost of 2 for the REMOTE operation reported in the body of the plan. Fortunately the SQL forwarded to the “remote” database doesn’t include this hint and (you’ll have to take my word for it) used an indexed access path into the table.

Where, though, is the indication that Oracle considered only 20 indexes? It’s in the 10053 trace file under the “Base Statistical Information” section in the subsection headed “Index Stats”:

Index Stats::
  Index: 0  Col#: 20    (NOT ANALYZED)
  LVLS: 1  #LB: 204  #DK: 250  LB/K: 1.00  DB/K: 400.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 19    (NOT ANALYZED)
  LVLS: 1  #LB: 204  #DK: 263  LB/K: 1.00  DB/K: 380.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 18    (NOT ANALYZED)
  LVLS: 1  #LB: 205  #DK: 277  LB/K: 1.00  DB/K: 361.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 17    (NOT ANALYZED)
  LVLS: 1  #LB: 205  #DK: 294  LB/K: 1.00  DB/K: 340.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 16    (NOT ANALYZED)
  LVLS: 1  #LB: 205  #DK: 312  LB/K: 1.00  DB/K: 320.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 15    (NOT ANALYZED)
  LVLS: 1  #LB: 205  #DK: 333  LB/K: 1.00  DB/K: 300.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 14    (NOT ANALYZED)
  LVLS: 1  #LB: 206  #DK: 357  LB/K: 1.00  DB/K: 280.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 13    (NOT ANALYZED)
  LVLS: 1  #LB: 206  #DK: 384  LB/K: 1.00  DB/K: 260.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 12    (NOT ANALYZED)
  LVLS: 1  #LB: 206  #DK: 416  LB/K: 1.00  DB/K: 240.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 11    (NOT ANALYZED)
  LVLS: 1  #LB: 206  #DK: 454  LB/K: 1.00  DB/K: 220.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 10    (NOT ANALYZED)
  LVLS: 1  #LB: 207  #DK: 500  LB/K: 1.00  DB/K: 200.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 9    (NOT ANALYZED)
  LVLS: 1  #LB: 207  #DK: 555  LB/K: 1.00  DB/K: 180.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 8    (NOT ANALYZED)
  LVLS: 1  #LB: 207  #DK: 625  LB/K: 1.00  DB/K: 160.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 7    (NOT ANALYZED)
  LVLS: 1  #LB: 208  #DK: 714  LB/K: 1.00  DB/K: 140.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 6    (NOT ANALYZED)
  LVLS: 1  #LB: 208  #DK: 833  LB/K: 1.00  DB/K: 120.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 5    (NOT ANALYZED)
  LVLS: 1  #LB: 208  #DK: 1000  LB/K: 1.00  DB/K: 100.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 4    (NOT ANALYZED)
  LVLS: 1  #LB: 208  #DK: 1250  LB/K: 1.00  DB/K: 80.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 3    (NOT ANALYZED)
  LVLS: 1  #LB: 209  #DK: 1666  LB/K: 1.00  DB/K: 60.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 2    (NOT ANALYZED)
  LVLS: 1  #LB: 209  #DK: 2500  LB/K: 1.00  DB/K: 40.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 
  Index: 0  Col#: 1    (NOT ANALYZED)
  LVLS: 1  #LB: 209  #DK: 5000  LB/K: 1.00  DB/K: 20.00  CLUF: 2002.00  NRW: 0.00 SSZ: 0.00 LGR: 0.00 CBK: 0.00 GQL: 0.00 CHR: 0.00 KQDFLG: 0 BSZ: 0
  KKEISFLG: 0 

We have 20 indexes listed, and while they’re all called “Index 0” (and reported as “Not Analyzed”) we can see from their column definitions that they are (in reverse order) the indexes on columns n01 through to n20 – i.e. the last 20 indexes created. The optimizer has created its plan based only on its knowledge of these indexes.

We might ask whether this matters or not – after all when the remote SQL gets to the remote database the remote optimizer is going to (re-)optimize it anyway and do the best it can with it, so at run-time Oracle could still end up using remote indexes that the local optimizer didn’t know about. So let’s get nasty and give the local optimizer a problem:

create index t2_id on t2(id);

select
        t1.id,
        t2.id,
        t2.padding
from
        t1                      t1,
        t2@&m_target            t2
where
        t1.id = 99
and     t2.n01 = t1.n01
;

I’ve created one more index on t2, which means the local optimizer is going to “forget” about the index that was the previous 20th index on the most recently created list for t2. That’s the index on (n01), which would have been a very good index for this query. If this query were to run locally the optimizer would do a nested loop from t1 to t2 using the index on (n01) – but the optimizer no longer knows about that index, so we get the following plan:

-------------------------------------------------------------------------------------------
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     | Inst   |IN-OUT|
-------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |      |       |       |   538 (100)|          |        |      |
|*  1 |  HASH JOIN         |      |    20 |  1140 |   538   (7)| 00:00:01 |        |      |
|*  2 |   TABLE ACCESS FULL| T1   |     1 |     9 |   268   (6)| 00:00:01 |        |      |
|   3 |   REMOTE           | T2   |   100K|  4687K|   268   (6)| 00:00:01 | ORCL@~ | R->S |
-------------------------------------------------------------------------------------------

Outline Data
-------------
  /*+
      BEGIN_OUTLINE_DATA
      IGNORE_OPTIM_EMBEDDED_HINTS
      OPTIMIZER_FEATURES_ENABLE('12.2.0.1')
      DB_VERSION('12.2.0.1')
      ALL_ROWS
      OUTLINE_LEAF(@"SEL$1")
      FULL(@"SEL$1" "T1"@"SEL$1")
      FULL(@"SEL$1" "T2"@"SEL$1")
      LEADING(@"SEL$1" "T1"@"SEL$1" "T2"@"SEL$1")
      USE_HASH(@"SEL$1" "T2"@"SEL$1")
      END_OUTLINE_DATA
  */

Predicate Information (identified by operation id):
---------------------------------------------------
   1 - access("T2"."N01"="T1"."N01")
   2 - filter("T1"."ID"=99)

Remote SQL Information (identified by operation id):
----------------------------------------------------
   3 - SELECT "N01","ID","PADDING" FROM "T2" "T2" (accessing 'ORCL@LOOPBACK' )

Oracle is going to do a hash join and apply the join predicate late. Although the remote optimizer can sometimes rescue us from a mistake made by the local optimizer and use indexes that the local optimizer doesn’t know about, there are times when the remote SQL generated by the local optimizer is so rigidly associated with the expected plan that there’s no way the remote optimizer can workaround the assumptions made by the local optimizer.

So when you create (or drop and recreate) an index, it’s just possible that a distributed plan will have to change because the local optimizer is no longer aware of an index that exists at the remote site.

tl;dr

Be very cautious about dropping and recreating indexes if the table in question

1. has more than 20 indexes
2. and is used at the remote end of a distributed execution plan

The optimizer will consider only 20 of the indexes on the table, choosing the ones with the highest object_ids. If you drop and recreate an index then it gets a new (highest) object_id and a plan may change because the index that Oracle was previously using is no longer in the top 20.

FBI Limitation

A recent question on the ODC (OTN) database forum prompted me to point out that the optimizer doesn’t consider function-based indexes on remote tables in distributed joins. I then spent 20 minutes trying to find the blog note where I had demonstrated this effect, or an entry in the manuals reporting the limitation – but I couldn’t find anything, so I’ve written a quick demo which I’ve run on 12.2.0.1 to show the effect. First, the SQL to create a couple of tables and a couple of indexes:

rem
rem     Script:         fbi_limitation.sql
rem     Author:         Jonathan Lewis
rem     Dated:          May 2018
rem

-- create public database link orcl@loopback using 'orcl'; 
define m_target = orcl@loopback

create table t1
segment creation immediate
nologging
as
with generator as (
        select
                rownum id
        from dual
        connect by
                level <= 1e4 -- > comment to avoid WordPress format issue
)
select
        rownum                          id,
        rownum                          n1,
        lpad(rownum,10,'0')             v1,
        lpad('x',100,'x')               padding
from
        generator       v1,
        generator       v2
where
        rownum <= 1e6 -- > comment to avoid WordPress format issue
;

create table t2
nologging
as
select * from t1
;

alter table t1 add constraint t1_pk primary key(id);
alter table t2 add constraint t2_pk primary key(id);
create unique index t2_f1 on t2(id+1);

begin
        dbms_stats.gather_table_stats(
                ownname     => user,
                tabname     => 'T1',
                cascade     => true,
                method_opt  => 'for all columns size 1'
        );

        dbms_stats.gather_table_stats(
                ownname     => user,
                tabname     => 'T2',
                cascade     => true,
                method_opt  => 'for all columns size 1'
        );
end;
/

The code is very simple, it creates a couple of identical tables with an id column that will produce an index with a very good clustering_factor. You’ll notice that I’ve (previously) created a public database link that is (in my case) a loopback to the current database and the code defines a variable that I can use as a substitution variable later on. If you want to do further tests with this model you’ll need to make some changes in these two lines.

So now I’m going to execute a query that should result in the optimizer choosing a nested loop between the tables – but I have two versions of the query, one which treats t2 as the local table it really is, and one that pretends (through the loopback) that t2 is remote.

set serveroutput off

select
        t1.v1, t2.v1
from
        t1,
        t2
--      t2@orcl@loopback
where
        t2.id+1 = t1.id
and     t1.n1 between 101 and 110
;


select * from table(dbms_xplan.display_cursor);

select
        t1.v1, t2.v1
from
        t1,
--      t2
        t2@orcl@loopback
where
        t2.id+1 = t1.id
and     t1.n1 between 101 and 110
;

select * from table(dbms_xplan.display_cursor);

Here are the two execution plans, pulled from memory – including the “remote” section in the distributed case:

SQL_ID  fthq1tqthq8js, child number 0
-------------------------------------
select  t1.v1, t2.v1 from  t1,  t2 -- t2@orcl@loopback where  t2.id+1 =
t1.id and t1.n1 between 101 and 110

Plan hash value: 1798294492

--------------------------------------------------------------------------------------
| Id  | Operation                    | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |       |       |       |  2347 (100)|          |
|   1 |  NESTED LOOPS                |       |    11 |   407 |  2347   (3)| 00:00:01 |
|*  2 |   TABLE ACCESS FULL          | T1    |    11 |   231 |  2325   (4)| 00:00:01 |
|   3 |   TABLE ACCESS BY INDEX ROWID| T2    |     1 |    16 |     2   (0)| 00:00:01 |
|*  4 |    INDEX UNIQUE SCAN         | T2_F1 |     1 |       |     1   (0)| 00:00:01 |
--------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   2 - filter(("T1"."N1"<=110 AND "T1"."N1">=101))
   4 - access("T2"."SYS_NC00005$"="T1"."ID")

Note
-----
   - this is an adaptive plan




SQL_ID  ftnmywddff1bb, child number 0
-------------------------------------
select  t1.v1, t2.v1 from  t1, -- t2  t2@orcl@loopback where  t2.id+1 =
t1.id and t1.n1 between 101 and 110

Plan hash value: 1770389500

-------------------------------------------------------------------------------------------
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     | Inst   |IN-OUT|
-------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |      |       |       |  4663 (100)|          |        |      |
|*  1 |  HASH JOIN         |      |    11 |   616 |  4663   (4)| 00:00:01 |        |      |
|*  2 |   TABLE ACCESS FULL| T1   |    11 |   231 |  2325   (4)| 00:00:01 |        |      |
|   3 |   REMOTE           | T2   |  1000K|    33M|  2319   (3)| 00:00:01 | ORCL@~ | R->S |
-------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------
   1 - access("T1"."ID"="T2"."ID"+1)
   2 - filter(("T1"."N1"<=110 AND "T1"."N1">=101))

Remote SQL Information (identified by operation id):
----------------------------------------------------
   3 - SELECT "ID","V1" FROM "T2" "T2" (accessing 'ORCL@LOOPBACK' )

Both plans show that the optimizer has estimated the number of rows that would be retrieved from t1 correctly (very nearly); but while the fully local query does a nested loop join using the high-precision, very efficient function-based index (reporting the internal supporting column referenced in the predicate section) the distributed query seems to have no idea about the remote function-based index and select all the required rows from the remote table and does a hash join.

Footnote:

Another reason for changes in execution plan when you test fully local and then run distributed is due to the optimizer ignoring remote histograms, as demonstrated in a much older blog note (though still true in 12.2.0.1).

Addendum

After finishing this note, I discovered that I had written a similar note about reverse key indexes nearly five years ago. Arguably a reverse key is just a special case of a function-based index – except it’s not labelled as such in user_tab_cols, and doesn’t depend on a system-generated hidden column.

 

Distributed Trap

Here’s an interesting (and potentially very useful) observation from an OTN database forum thread that appeared at the end of last week. It concerns the problems of pulling data from remote systems, and is an interesting example because it highlights a case where you might want to ignore the normal performance penalties associated with row-level triggers. I’ll start by building some data:

rem
rem     Script:         remote_insert_2.sql
rem     Author:         Jonathan Lewis
rem     Dated:          Nov 2016
rem
rem     Last tested
rem             12.1.0.2
rem             11.2.0.4
rem

create table t1
as
with generator as (
        select
                rownum id
        from dual
        connect by
                level <= 1e4
)
select
        rownum                  id,
        lpad(rownum,10,'0')     small_vc,
        rpad('x',100)           padding
from
        generator       v1,
        generator       v2
where
        rownum <= 50000
;

alter table t1 add constraint t1_pk primary key(id);

create table t2
as
with generator as (
        select
                rownum id
        from dual
        connect by
                level <= 1e4
)
select
        trunc(sysdate) + rownum d1,
        rownum                  id,
        lpad(rownum,10,'0')     small_vc,
        rpad('x',50)            padding
from
        generator
where
        rownum <= 500
;

alter table t2 add constraint t2_pk primary key(id);

create table t3
as
select sysdate d1, t1.* from t1
where rownum = 0
;

--  Now gather stats if you need to (depending on version)

I’ve created three tables. Table t3 is an empty copy of table t1 with a date column added, and t2 is some sort of reference table that looks rather like table t1 but has a lot less data. Now I’m going to pretend that t1 and t2 are in a remote database while t3 is in the local database and copy data from t1 to t3, omitting any data that is referenced in t2. The SQL is simple:

define m_target=test@loopback

insert into t3(
        id, small_vc, padding
)
select
        t1.id, t1.small_vc, t1.padding
from
        t1@&m_target    t1
where
        t1.id not in (
                select t2.id from t2@&m_target
        )
;

----------------------------------------------------------------------
| Id  | Operation                | Name | Cost (%CPU)| Inst   |IN-OUT|
----------------------------------------------------------------------
|   0 | INSERT STATEMENT         |      |     0   (0)|        |      |
|   1 |  LOAD TABLE CONVENTIONAL | T3   |            |        |      |
|   2 |   REMOTE                 |      |            |   TEST | R->S |
----------------------------------------------------------------------

Remote SQL Information (identified by operation id):
----------------------------------------------------
   2 - EXPLAIN PLAN INTO PLAN_TABLE@! FOR SELECT
       "A1"."ID","A1"."SMALL_VC","A1"."PADDING" FROM "T1" "A1" WHERE
       "A1"."ID"<>ALL (SELECT "A2"."ID" FROM "T2" "A2") (accessing
       'TEST.LOCALDOMAIN@LOOPBACK' )

I’ve set up an SQL*Plus substitution variable to hold a database link name (and used a loopback qualifier so that I can pretend t1 and t2 are in a remote database. The execution plan (taken from an explain plan, but confirmed by running the query and calling dbms_xplan.display_cursor) shows that Oracle has executed select part of the insert as a “fully remote” statement – which is nice.

Unfortunately I forgot to include a datestamp as I copied the data over. So let’s modify the query to do that:

insert into t3(
        d1,
        id, small_vc, padding
)
select
        sysdate,
        t1.id, t1.small_vc, t1.padding
from
        t1@&m_target    t1
where
        t1.id not in (
                select t2.id from t2@&m_target
        )
;

--------------------------------------------------------------------------------------------------------
| Id  | Operation                | Name | Starts | E-Rows | Cost (%CPU)| A-Rows |   A-Time   | Buffers |
--------------------------------------------------------------------------------------------------------
|   0 | INSERT STATEMENT         |      |      1 |        |   123 (100)|      0 |00:00:27.42 |   10908 |
|   1 |  LOAD TABLE CONVENTIONAL |      |      1 |        |            |      0 |00:00:27.42 |   10908 |
|*  2 |   FILTER                 |      |      1 |        |            |  49500 |00:00:26.51 |       0 |
|   3 |    REMOTE                | T1   |      1 |  50000 |   113   (3)|  50000 |00:00:00.33 |       0 |
|   4 |    REMOTE                | T2   |  50000 |      1 |     0   (0)|    500 |00:00:25.29 |       0 |
--------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------
   2 - filter( NOT EXISTS (SELECT 0 FROM  "T2" WHERE "T2"."ID"=:B1))

Remote SQL Information (identified by operation id):
----------------------------------------------------
   3 - SELECT /*+ OPAQUE_TRANSFORM */ "ID","SMALL_VC","PADDING" FROM "T1" "T1"
       (accessing 'TEST.LOCALDOMAIN@LOOPBACK' )

   4 - SELECT /*+ OPAQUE_TRANSFORM */ "ID" FROM "T2" "T2" WHERE "ID"=:1 (accessing
       'TEST.LOCALDOMAIN@LOOPBACK' )

Whoops, the plan just changed – it also took 27.4 seconds instead of the 1.1 seconds that it used to – that’s because of the 50,000 remote calls to execute a remote query for the subquery filter. The addition of sysdate (which is the local sysdate@!) to the select list has made the select statement distributed instead of fully remote, and the query for a CTAS or “insert/select” has to be driven from the local site if it’s a distributed query.

Okay, plan (b), don’t insert sysdate, add it to the table as a default:

alter table t3 modify d1 default sysdate;

This doesn’t help; even though the query doesn’t mention sysdate explicitly the query is still treated as disrtibuted query.

Okay, plan (c) – don’t insert sysdate, add a trigger to the table:

alter table t3 modify d1 default null;

create or replace trigger t3_bri
before insert on t3
for each row
begin
        :new.d1 := sysdate;
end;
/

And this works (in 11.2.0.4 and 12.1.0.2, at least, which are the versions I tested).

I could have made the trigger a little more sophisticated, of course, but the point of the post was simply to demonstrate a problem and a simple workaround.

There are probably several other commonly used features (various sys_context() calls, perhaps) that have the same effect.

 

Uniquely parallel

Here’s a surprising (to me) execution plan from 12.1.0.2 – parallel execution to find one row in a table using a unique scan of a unique index – produced by running the following script (data creation SQL to follow):

set serveroutput off
set linesize 180
set trimspool on
set pagesize 60

alter session set statistics_level = all;

variable b1 number
exec :b1 := 50000

select /*+ parallel (3) */ id, v1 from t2 where id=:b1;

select * from table(dbms_xplan.display_cursor(null,null,'allstats parallel'));

break on dfo_number skip 1 on tq_id skip 1 on server_type

select
        dfo_number, tq_id, server_type, instance, process, num_rows
from
        v$pq_tqstat
order by
        dfo_number, tq_id, server_type desc, instance, process
;

All I’ve done is enable rowsource execution statistics, set a bind variable to a value, query a table with a /*+ parallel(3) */ hint to find the one row that will be identified by primary key, and then reported the actual execution plan. When I first ran the test Oracle didn’t report the execution statistics correctly so I’ve also queried v$pq_tqstat to show the PX servers used and the flow of data through the plan. Here’s the plan, followed by the  results from v$pq_tqstat:

SQL_ID  0dzynh9d29pt9, child number 0
-------------------------------------
select /*+ parallel (3) */ id,v1 from t2 where id=:b1

Plan hash value: 247082613

---------------------------------------------------------------------------------------------------------------------------
| Id  | Operation                         | Name     | Starts | E-Rows |    TQ  |IN-OUT| PQ Distrib | A-Rows |   A-Time   |
---------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                  |          |      1 |        |        |      |            |      1 |00:00:00.02 |
|   1 |  PX COORDINATOR                   |          |      1 |        |        |      |            |      1 |00:00:00.02 |
|   2 |   PX SEND QC (RANDOM)             | :TQ10001 |      0 |      1 |  Q1,01 | P->S | QC (RAND)  |      0 |00:00:00.01 |
|   3 |    TABLE ACCESS BY INDEX ROWID    | T2       |      0 |      1 |  Q1,01 | PCWP |            |      0 |00:00:00.01 |
|   4 |     BUFFER SORT                   |          |      0 |        |  Q1,01 | PCWC |            |      0 |00:00:00.01 |
|   5 |      PX RECEIVE                   |          |      0 |      1 |  Q1,01 | PCWP |            |      0 |00:00:00.01 |
|   6 |       PX SEND HASH (BLOCK ADDRESS)| :TQ10000 |      0 |      1 |  Q1,00 | S->P | HASH (BLOCK|      0 |00:00:00.01 |
|   7 |        PX SELECTOR                |          |      0 |        |  Q1,00 | SCWC |            |      0 |00:00:00.01 |
|*  8 |         INDEX UNIQUE SCAN         | T2_PK    |      0 |      1 |  Q1,00 | SCWP |            |      0 |00:00:00.01 |
---------------------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------
   8 - access("ID"=:B1)

Note
-----
   - Degree of Parallelism is 3 because of hint

DFO_NUMBER      TQ_ID SERVER_TYP   INSTANCE PROCES   NUM_ROWS
---------- ---------- ---------- ---------- ------ ----------
         1          0 Producer            1 P003            0
                                          1 P004            1
                                          1 P005            0
                      Consumer            1 P000            0
                                          1 P001            1
                                          1 P002            0

                    1 Producer            1 P000            0
                                          1 P001            1
                                          1 P002            0
                      Consumer            1 QC              1

As you can see the table access follows a unique scan of an index and, although the rowsource execution stats report zero starts for the unique scan, we can see from v$pq_tqstat that slave P004 acquired a “row” (actually a rowid) and passed it to slave P001 which then acquired a row from the table and passed that row to the query coordinator. Oracle really did execute a parallel query, starting and stopping a total of 6 sessions to perform a single unique index access.

You’ll notice operation 7 is one you’ve only seen in the latest version of Oracle. The PX SELECTOR was introduced in 12c to reduce the number of times a complex parallel query would funnel into the query coordinator (parallel to serial) and then fan out again (serial to parallel) generating a new data flow operation tree (DFO tree) spawning one or two new parallel server groups as it did so. To stop this happening a step that needs to serialise in a 12c parallel plan can nominate one of the existing PX server processes (from each set, if necessary) to do the job so that the same set of PX servers can carry on running the query without the need for a new DFO tree to appear.

This enhancement to parallel execution plans is a good idea – except when it appears in my silly little query and turns something that ought to be quick and cheap into a job that is far more resource-intensive than it should be.

At this point, of course, you’re probably wondering what kind of idiot would put a parallel() hint into a query that was doing nothing but selecting one row by primary key – the answer is: “the Oracle optimizer in 12c”. I discovered this anomaly while creating a demonstration of the way that a distributed parallel query has to serialise through a single database link even if the operations at the two ends of the link run parallel. Here’s the SQL I wrote for the full demonstration:

rem     Script:         distributed_pq.sql
rem     Author:         Jonathan Lewis
rem     Dated:          May 2016

define m_remote='test@loopback'
define m_remote='orcl@loopback'

create table t1
nologging
as
with generator as (
        select  --+ materialize
                rownum id 
        from dual 
        connect by 
                level <= 1e4    -- > comment to avoid wordpress format issue
)
select
        cast(rownum as number(8,0))                     id,
        cast(lpad(rownum,8,'0') as varchar2(8))         v1,
        cast(rpad('x',100) as varchar2(100))            padding
from
        generator       v1,
        generator       v2
where
        rownum <= 1e5    -- > comment to avoid wordpress format issue
;

create table t2
as
select  *
from    t1
where   mod(id,100) = 0
;

alter table t2 add constraint t2_pk primary key(id);

begin
        dbms_stats.gather_table_stats(
                ownname          => user,
                tabname          =>'T1',
                method_opt       => 'for all columns size 1'
        );

        dbms_stats.gather_table_stats(
                ownname          => user,
                tabname          =>'T2',
                method_opt       => 'for all columns size 1'
        );
end;
/


set serveroutput off

select
        /*+ parallel(3) */
        t1.v1, t2.v1
from
        t1,
        t2@&m_remote
where
        mod(t1.id,10) = 0
and     t2.id = t1.id
and     mod(to_number(t2.v1),10) = 1
;

select * from table(dbms_xplan.display_cursor(null,null,'allstats parallel'));

If you want to run this demo you’ll need to do something about formatting the output; more importantly you’ll have to create a database link (with a loopback link) and set up a define identifying it at the line where I’ve got orcl@loopback and test@loopback (which are my 12c and 11g loopback links respectively).

Here’s the plan (with rowsource stats) I got from the 12c test:

----------------------------------------------------------------------------------------------------------------------------------
| Id  | Operation             | Name     | Starts | E-Rows |    TQ  |IN-OUT| PQ Distrib | A-Rows |   A-Time   | Buffers | Reads  |
----------------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |          |      1 |        |        |      |            |      0 |00:01:14.67 |       7 |      0 |
|   1 |  NESTED LOOPS         |          |      1 |     10 |        |      |            |      0 |00:01:14.67 |       7 |      0 |
|   2 |   PX COORDINATOR      |          |      1 |        |        |      |            |  10000 |00:00:00.11 |       7 |      0 |
|   3 |    PX SEND QC (RANDOM)| :TQ10000 |      0 |   1000 |  Q1,00 | P->S | QC (RAND)  |      0 |00:00:00.01 |       0 |      0 |
|   4 |     PX BLOCK ITERATOR |          |      3 |   1000 |  Q1,00 | PCWC |            |  10000 |00:03:17.72 |    1745 |   1667 |
|*  5 |      TABLE ACCESS FULL| T1       |     39 |   1000 |  Q1,00 | PCWP |            |  10000 |00:00:00.06 |    1745 |   1667 |
|   6 |   REMOTE              | T2       |  10000 |      1 |        |      |            |      0 |00:01:14.44 |       0 |      0 |
----------------------------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------
   5 - access(:Z>=:Z AND :Z<=:Z)
       filter(MOD("T1"."ID",10)=0)

Remote SQL Information (identified by operation id):
----------------------------------------------------
   6 - SELECT /*+ SHARED (3) */ "ID","V1" FROM "T2" "T2" WHERE "ID"=:1 AND MOD(TO_NUMBER("V1"),10)=1
       (accessing 'ORCL@LOOPBACK' )

Note
-----
   - Degree of Parallelism is 3 because of hint

I have hacked this output a little – the “Remote SQL” section didn’t get reported by display_cursor(), so I’ve inserted the remote sql I got from a call to dbms_xplan.display() after using explain plan to generate a plan. Note the /*+ shared(3) */ hint that appears in the remote SQL – that’s the internal version of a parallel(3) hint.

In 11g the query complete in 2.4 seconds, in 12c the query took nearly 75 seconds to run thanks to the 12c enhancement that allowed it to obey the hint! Looking at the time column (and ignoring the anomalous 3:17 at operation 4 – which might roughly be echoing 3 * 1:14) we can see that the time goes on the calls to the remote database (and a check of v$session_event shows this time spent in “SQL*Net message from db link”), so the obvious thing to do is check what actually happened at the remote database and we can do that by searching the library cache for a recognizable piece of the remote SQL – here’s the SQL to do that, with the results from 11g followed by the results from 12c:

SQL> select sql_id, child_number, executions, px_servers_executions, sql_text from v$sql
  2  where sql_text like '%SHARED%' and sql_text not like 'select sql_id%';

11g results
SQL_ID        CHILD_NUMBER EXECUTIONS PX_SERVERS_EXECUTIONS
------------- ------------ ---------- ---------------------
SQL_TEXT
------------------------------------------------------------------------------------------------------------------------------------
c0f292z5czhwk            0      10000                     0
SELECT /*+ SHARED (3) */ "ID","V1" FROM "T2" "T2" WHERE MOD(TO_NUMBER("V1"),10)=1 AND "ID"=:1

12c results
SQL_ID        CHILD_NUMBER EXECUTIONS PX_SERVERS_EXECUTIONS
------------- ------------ ---------- ---------------------
SQL_TEXT
------------------------------------------------------------------------------------------------------------------------------------
7bk51w7vtagwd            0      10000                     0
SELECT /*+ SHARED (3) */ "ID","V1" FROM "T2" "T2" WHERE "ID"=:1 AND MOD(TO_NUMBER("V1"),10)=1

7bk51w7vtagwd            1          0                 59995
SELECT /*+ SHARED (3) */ "ID","V1" FROM "T2" "T2" WHERE "ID"=:1 AND MOD(TO_NUMBER("V1"),10)=1

It’s not surprising to see that the query has executed 10,000 times – that’s what we were told by the Starts statistic from dbms_output.display_cursor(), but 12c has 60,000 (with a little error) PX Servers executions of the statement. That’s 10,000 executions * degree 3 * the 2 slave sets we saw in my original execution plan. (It’s an odd little quirk of the two versions of Oracle that the order of predicates in the remote SQL was reversed between 11g and 12c – leading to two different SQL_IDs).

By enabling rowsource execution stats at the system level I was able to capture the remote execution plan with its stats:

SQL_ID  7bk51w7vtagwd, child number 0
-------------------------------------
SELECT /*+ SHARED (3) */ "ID","V1" FROM "T2" "T2" WHERE "ID"=:1 AND
MOD(TO_NUMBER("V1"),10)=1

--------------------------------------------------------------------------------------------------------
| Id  | Operation                         | Name     | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
--------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                  |          |      0 |        |      0 |00:00:00.01 |       0 |
|   1 |  PX COORDINATOR                   |          |      0 |        |      0 |00:00:00.01 |       0 |
|   2 |   PX SEND QC (RANDOM)             | :TQ10001 |      0 |      1 |      0 |00:00:00.01 |       0 |
|*  3 |    TABLE ACCESS BY INDEX ROWID    | T2       |  29983 |      1 |      0 |00:00:22.21 |    1000 |
|   4 |     BUFFER SORT                   |          |  29995 |        |    999 |00:00:21.78 |       0 |
|   5 |      PX RECEIVE                   |          |  29924 |      1 |    994 |00:00:21.21 |       0 |
|   6 |       PX SEND HASH (BLOCK ADDRESS)| :TQ10000 |      0 |      1 |      0 |00:00:00.01 |       0 |
|   7 |        PX SELECTOR                |          |  29993 |        |    999 |00:00:06.08 |   19992 |
|*  8 |         INDEX UNIQUE SCAN         | T2_PK    |  29999 |      1 |   1000 |00:00:00.24 |   20000 |
--------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------
   3 - filter(MOD(TO_NUMBER("V1"),10)=1)
   8 - access("ID"=:1)

Unlike the test case I started with, this output did show the number of starts (with a few missing) and the work done across the slaves. Our index probe had to do two buffer gets on every execution, and we have 10,000 executions of the query so 20,000 buffer gets on the index unique scan. Even though only one slave actually does any work with the PX Selector, all three slaves in that set seem to “start” the relevant operations. The definition of the data meant that only one index probe in 10 was successful so we only visited 1,000 buffers from the table. If you’re familiar with reading plans with rowsource execution stats you’ll appreciate that something has gone wrong in the reporting here – that 1,000 at operation 3 should read 21,000 because it ought to include the 20,000 from the index scan (at least, that’s what a serial plan would do).

If you’re still wondering why running this query as a parallel query should take so long – after all it’s only 10,000 executions in 70 seconds – bear in mind that Oracle has to allocate and deallocate 6 PX servers to new sessions each time it starts; the instance activity stats showed “logons cumulative” going up by 60,000 each time I ran the driving query: that’s about 850 logons (and log offs) per second. I don’t think my test machine would give a realistic impression of the impact of a couple of copies of this query running simultaneously, but when I tried the contention introduce increased the run time to 93 seconds.

tl;dr

Watch out for poor performance becoming disastrous for distributed parallel queries when you upgrade from 11g to 12c

Update (May 2017)

Thanks to Mauro Pagano – identifying this as fix 13345888 (check v$system_fix control) introduced in 12.1.0.1

Distributed Sets

In an earlier post I’ve described how a distributed query can operate at a remote site if it’s a simple select but has to operate at the local site if it’s a CTAS (create as select) or insert as select. There’s (at least) one special case where this turns out to be untrue … provided you write the query in the correct fashion. I discovered this only as a result of doing a few experiments in response to a question on the OTN database forum.

Distributed Queries – 3

A comment I’ve made many times in the past about distributed queries is that Oracle doesn’t try to retrieve histogram information from remote databases when optimizing a query. Checking back through previous posts, though, I don’t think I’ve ever mentioned it on the blog – so here’s an example demonstrating the point.

Distributed Queries – 2

I have often said that the optimizer “forgets” that it is dealing with a distributed query once it has collected the stats that it can about the objects in the query, and that as a consequence the driving site for a distributed query will be the local database unless you use the /*+ driving_site */ hint to change it.

While investigating an oddity with a distributed query between two 11.1.0.7 databases a few days, I noticed something in the 10053 trace file that made me change my mind, and go back to look at earlier versions of Oracle.

Distributed Objects

I recently came across a tidy solution to a common problem – how to minimise code maintenance in a procedure while maximising flexibility of the procedure. The task was fairly simple – create a ref cursor for a calling program to return data that (a) followed complex selection rules and (b) allowed the user to specify numerous types of input.

The principle was simple – the final ref cursor was driven by a list of (say) order ids – and the details to be returned about those orders required some fairly complex SQL to execute. To separate the complexity of constructing the list of columns from the complexity of identifying the required rows the developers had split the procedure into two stages. First, select the list of relevant order ids using one of several possible statements – the appropriate statement being derived from analysis of the inputs to the procedure; secondly open a ref cursor using that list of order ids. In this way if a new set of rules for selection appeared the only new code needed was a new query to select the ids – the main body of code didn’t need to be modified and re-optimised.

Distributed Pipelines

In an article that I wrote about the  driving_site() hint a few months ago I pointed out that the hint was not supposed to work with “create table as select” (CTAS) and “insert as select”. One of the people commenting on the note mentioned pipelined function as a workaround to this limitation – and I’ve finally got around to writing a note about the method.

The idea is simple. If you can write a distributed select statement that takes advantage of the /*+ driving_site(alias) */ hint to work efficiently, you can wrap the statement in a pl/sql cursor loop and stick that loop into a pipelined function to maximise the efficiency of create or insert as select. Here’s some sample code (tested on 11.1.0.6) to demonstrate the principle:

Ignoring Hints

I’ve previously published a couple of notes (hereand here) about the use of the driving_site() hint with distributed queries. The first note pointed out that the hint was deliberately ignored if you write a local CTAS or INSERT that did a distributed query. I’ve just found another case where the hint is ignored – this time in a simple SELECT statement.

Try running an ordinary distributed query from the SYS account, and then try using the driving_site() hint to make it run at the remote site. When I tried this a few days ago I ended up wasting half an hour translating some SQL from ANSI to Oracle dialect because I thought that the ANSI was making Oracle transform the query in a way that lost the hint – then I discovered that both versions of the code worked correctly if I logged in as a different user.

I was running my queries between two databases using 11.1.0.7 – I won’t guarantee you get the same results on other versions, but it looks like SYS doesn’t honour the driving_site() hint. I can’t think of a robust argument why this should be the case, but if I were forced to do some vague hand-waving I’d probably mumble something about potential security loopholes.

Footnote: I should, of course, have mentioned that there are all sorts of things that behave in unexpected ways if you are logged on as SYS, and that you shouldn’t be logged on as SYS – especially in a production system.

[Further reading on “ignoring hints”]

Distributed Queries

Some time ago I wrote a note about distributed DML, pointing out that the driving_site() hint works with distributed queries but not with DML based on distributed queries; so insert as select, or create as select and so on will “ignore” the hint.

This is just a little follow-up to give you an  idea of what execution plans for distributed queries look like so that you can tell whether your query is going to work locally or remotely.

Distributed DML

Someone recently sent me a request about a piece of SQL they could not optimise. I don’t usually respond to private requests – it’s not an effective use of my time – but their example was something that pops up relatively frequently as a “bug” – so I thought I’d mention it here.

The SQL looked like this:

insert into tab3
select 				-- small result set
	*
from
	tab1@dblink	t1	-- large data set
where
	tab1.col1 in (
		select
			col1
		from
			tab2	-- small data set
	)

Distributed Oddity

An interesting side-effect of blogging (rather than writing up a more formal document for my web site) is that it feels perfectly reasonable to throw out some odd observations without necessarily having to supply an explanation or solution.

So here’s the bare bones of an oddity relating to distributed queries that someone emailed me a couple of days ago:

INST1 -  9.2.0.6:  Insert data into table X and commit 
INST2 - 10.2.0.1:  select from X@DB1,dual;