Thursday, November 17, 2016

Run time SQL Statistics in Production with SQL Patch

Sometimes we need to “test” in production.  Or maybe we just want to see more statistics of a SQL in production.  To get full good statistics on a SQL statement running in Oracle you need one of two things, STATISTICS_LEVEL set to ALL or use the GATHER_PLAN_STATISTICS hint in the statement.

Cool as STATISTICS_LEVEL to ALL is, this is likely not practical in most production environments.  It adds a level of work that likely will be noticed.  The reason is quite simple, to get accurate timing statistics in particular Oracle polls the system clock at a very high rate and this added to the run time of the SQL.  And this will happen to all SQL on the system, making everything slow down.  Not something most folks want.

Using the hint is nice because it focuses this extra work to the statement and only the statement.  But how do you get the hint into already existing code?  And what if you don’t have access to the source?  And even if you do, if you add a hint, you’d like have to go thru change control to punch it out to your production system. Ugh.

Enter SQL PATCH. 

Starting with Oracle 11 this functionality gives you the ability to add a hint to a SQL ID without modifying the code.  This is pretty cool, but I’d like to caution you that this shouldn’t be used often. And when used it should be a temporary fix.  And this case of getting run time statistics is a good example of how it could be used in that context.

The basic steps are:
1.  Identify the SQL you want to gather statistics on.
2.   Create a patch for the SQL with 'GATHER_PLAN_STATISTICS' as the hint to be added with DBMS_SQLDIAG_INTERNAL.I_CREATE_PATCH.  When a patch is created it is enabled by default.   NOTE: to create a patch you must be connected as SYSDBA.
3.  Then either disable the patch or drop it once you’re done.

Also I’d like give some kudos to Shiraz Kamal who asked me about doing this in a recent class which got me to write the demo and this blog post.  Thanks Z!

Here is an example:

SQL>
SQL> @sql_patch_demo2
SQL> -- sql_patch_demo2.sql
SQL> -- RVD November 2016
SQL> --
SQL>
SQL> set pages 9999
SQL> set lines 150
SQL> column plan_table_output format a130
SQL> column NAME format a20
SQL> column CATEGORY format a20
SQL> column STATUS format a10
SQL>
SQL> set echo on
SQL>
SQL> DECLARE
  2    PATCHNOTFND EXCEPTION;
  3    PRAGMA EXCEPTION_INIT(PATCHNOTFND, -13833);
  4  BEGIN
  5    DBMS_SQLDIAG.DROP_SQL_PATCH('TEST_PATCH99');
  6  EXCEPTION
  7    WHEN PATCHNOTFND THEN
  8      dbms_output.put_line('*** Patch Not Found ***');
  9  END;
 10  /
*** Patch Not Found ***
SQL>
SQL> select name,category,status,sql_text from dba_sql_patches;
SQL>
SQL> @hflush sp
SQL> set echo off
*** Shared pool flushed ***

PL/SQL procedure successfully completed.

SQL>
SQL> alter session set statistics_level=typical
  2  /
SQL>
SQL> set serveroutput off
SQL> set termout off
SQL>
SQL> select * from table(dbms_xplan.display_cursor(format=>'allstats last'));

PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  d1v4v24j9y297, child number 0
-------------------------------------
select ord2.order_no, cust_no, order_date, product_id, quantity,
item_price from ord_item2, ord2 where ord_item2.order_no =
ord2.order_no and ord2.order_no between 10000 and 20000

Plan hash value: 2269185734

----------------------------------------------------------------------------
| Id  | Operation          | Name      | E-Rows |  OMem |  1Mem | Used-Mem |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |        |       |       |          |
|*  1 |  HASH JOIN         |           |    134 |  1263K|  1263K| 1245K (0)|
|*  2 |   TABLE ACCESS FULL| ORD2      |    131 |       |       |          |
|*  3 |   TABLE ACCESS FULL| ORD_ITEM2 |    721 |       |       |          |
----------------------------------------------------------------------------

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

   1 - access("ORD_ITEM2"."ORDER_NO"="ORD2"."ORDER_NO")
   2 - filter(("ORD2"."ORDER_NO"<=20000 AND "ORD2"."ORDER_NO">=10000))
   3 - filter(("ORD_ITEM2"."ORDER_NO"<=20000 AND
              "ORD_ITEM2"."ORDER_NO">=10000))

Note
-----
   - this is an adaptive plan
   - Warning: basic plan statistics not available. These are only collected when:
       * hint 'gather_plan_statistics' is used for the statement or
       * parameter 'statistics_level' is set to 'ALL', at session or system level

SQL>
SQL> set serveroutput on
SQL> -- ***************************************************************************
SQL> -- Notice the message that only basic stats are available
SQL> -- Now to create a patch to give us the stats we want
SQL> -- Need to connect as SYS to create the patch
SQL> -- ***************************************************************************
SQL>
SQL> connect / as sysdba
Connected.
SQL>
SQL> BEGIN
  2    SYS.DBMS_SQLDIAG_INTERNAL.I_CREATE_PATCH(
  3        SQL_TEXT  => 'select ord2.order_no, cust_no, order_date, product_id, quantity,'||
  4                     'item_price from ord_item2, ord2 where ord_item2.order_no ='||
  5                     'ord2.order_no and ord2.order_no between 10000 and 20000',
  6        HINT_TEXT => 'GATHER_PLAN_STATISTICS',
  7        NAME      => 'TEST_PATCH99');
  8  END;
  9  /
SQL>
SQL> -- ***************************************************************************
SQL> -- Reconnect as OP
SQL> -- ***************************************************************************
SQL> connect op/op
Connected.
SQL>
SQL> select name,category,status,sql_text from dba_sql_patches;

NAME                 CATEGORY             STATUS     SQL_TEXT
-------------------- -------------------- ---------- --------------------------------------------------------------------------------
TEST_PATCH99         DEFAULT              ENABLED    select ord2.order_no, cust_no, order_date, product_id, quantity,item_price from
SQL>
SQL> -- ***************************************************************************
SQL> -- Lets see if the SQL patch has an effect
SQL> -- ***************************************************************************
SQL>
SQL> set serveroutput off
SQL> set termout off
SQL>
SQL> select * from table(dbms_xplan.display_cursor(format=>'allstats last'));

PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------------------------------------------------------------
SQL_ID  d1v4v24j9y297, child number 0
-------------------------------------
select ord2.order_no, cust_no, order_date, product_id, quantity,
item_price from ord_item2, ord2 where ord_item2.order_no =
ord2.order_no and ord2.order_no between 10000 and 20000

Plan hash value: 2269185734

---------------------------------------------------------------------------------------------------------------------
| Id  | Operation          | Name      | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
---------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |      1 |        |    723 |00:00:00.01 |     557 |       |       |          |
|*  1 |  HASH JOIN         |           |      1 |    134 |    723 |00:00:00.01 |     557 |  1263K|  1263K| 1310K (0)|
|*  2 |   TABLE ACCESS FULL| ORD2      |      1 |    131 |    132 |00:00:00.01 |     191 |       |       |          |
|*  3 |   TABLE ACCESS FULL| ORD_ITEM2 |      1 |    721 |    723 |00:00:00.01 |     366 |       |       |          |
---------------------------------------------------------------------------------------------------------------------

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

   1 - access("ORD_ITEM2"."ORDER_NO"="ORD2"."ORDER_NO")
   2 - filter(("ORD2"."ORDER_NO"<=20000 AND "ORD2"."ORDER_NO">=10000))
   3 - filter(("ORD_ITEM2"."ORDER_NO"<=20000 AND "ORD_ITEM2"."ORDER_NO">=10000))

Note
-----
   - SQL patch "TEST_PATCH99" used for this statement
   - this is an adaptive plan

SQL>
SQL>
SQL> -- ***************************************************************************
SQL> -- Disable the patch and go again
SQL> -- ***************************************************************************
SQL>
SQL>
SQL> EXEC DBMS_SQLDIAG.ALTER_SQL_PATCH('TEST_PATCH99', 'STATUS', 'DISABLED');
SQL>
SQL> select name,category,status,sql_text from dba_sql_patches;

NAME                 CATEGORY             STATUS     SQL_TEXT
-------------------- -------------------- ---------- --------------------------------------------------------------------------------
TEST_PATCH99         DEFAULT              DISABLED   select ord2.order_no, cust_no, order_date, product_id, quantity,item_price from
SQL>
SQL> set serveroutput off
SQL> set termout off
SQL>
SQL> select * from table(dbms_xplan.display_cursor(format=>'allstats last'));

PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------------------------------------------------------------
SQL_ID  d1v4v24j9y297, child number 0
-------------------------------------
select ord2.order_no, cust_no, order_date, product_id, quantity,
item_price from ord_item2, ord2 where ord_item2.order_no =
ord2.order_no and ord2.order_no between 10000 and 20000

Plan hash value: 2269185734

---------------------------------------------------------------------------------------------------------------------
| Id  | Operation          | Name      | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
---------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |      1 |        |    723 |00:00:00.01 |     557 |       |       |          |
|*  1 |  HASH JOIN         |           |      1 |    134 |    723 |00:00:00.01 |     557 |  1263K|  1263K| 1295K (0)|
|*  2 |   TABLE ACCESS FULL| ORD2      |      1 |    131 |    132 |00:00:00.01 |     191 |       |       |          |
|*  3 |   TABLE ACCESS FULL| ORD_ITEM2 |      1 |    721 |    723 |00:00:00.01 |     366 |       |       |          |
---------------------------------------------------------------------------------------------------------------------

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

   1 - access("ORD_ITEM2"."ORDER_NO"="ORD2"."ORDER_NO")
   2 - filter(("ORD2"."ORDER_NO"<=20000 AND "ORD2"."ORDER_NO">=10000))
   3 - filter(("ORD_ITEM2"."ORDER_NO"<=20000 AND "ORD_ITEM2"."ORDER_NO">=10000))

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

SQL>
SQL> -- ***************************************************************************
SQL> -- Notice that the patch is not used. The patch is not mentioned in the notes.
SQL> -- The plans still has stats from the prior run which are shown.
SQL> -- ***************************************************************************
SQL>
SQL>
SQL> set serveroutput on
SQL> -- ***************************************************************************
SQL> -- Now time to clean up
SQL> -- ***************************************************************************
SQL>
SQL> DECLARE
  2    PATCHNOTFND EXCEPTION;
  3    PRAGMA EXCEPTION_INIT(PATCHNOTFND, -13833);
  4  BEGIN
  5    DBMS_SQLDIAG.DROP_SQL_PATCH('TEST_PATCH99');
  6  EXCEPTION
  7    WHEN PATCHNOTFND THEN
  8      dbms_output.put_line('*** Patch Not Found ***');
  9  END;
 10  /
SQL>
SQL> select name,category,created, sql_text from dba_sql_patches;
SQL>

4 comments:

  1. Hm, why not use sql monitor as it provides more data than display_cursor and is automatically gathered for queries taking >=5s.

    ReplyDelete
  2. First it requires Enterprise Edition and the Diagnostics and Tuning option, I try to show examples that anyone can use without any "extra" licensing. Second it's easier to use DISPLAY_CURSOR in SQL Plus. Yes you can use the DBMS_SQLTUNE package but you still need to be licensed for it. If you have all that then sure, use it!

    ReplyDelete
  3. Agreed, might be helpful to have that goal explicitly stated in the blog post :)

    ReplyDelete
  4. This comment has been removed by a blog administrator.

    ReplyDelete