Our monitoring software found a long running job on one of our client’s databases. The job was an Oracle’s auto task running statistics gathering for more than 3 hours. I was curious to know why it took so long and used a query to ASH to find out the most common SQL during the job run based on the module name. Results were surprising to me: top SQL was coming with SQL Tuning Advisor comment.
Here is the SQL I used:
SQL> select s.sql_id, t.sql_text, s.cnt 2 from 3 (select * 4 from 5 ( 6 select sql_id, count(*) cnt 7 from v$active_session_history 8 where action like 'ORA$AT_OS_OPT_SY%' 9 group by sql_id 10 order by count(*) desc 11 ) 12 where rownum <= 5 13 ) s, 14 dba_hist_sqltext t 15 where s.sql_id = t.sql_id; SQL_ID SQL_TEXT CNT ------------- -------------------------------------------------------------------------------- ---------- 020t65s3ah2pq select substrb(dump(val,16,0,32),1,120) ep, cnt from (select /*+ no_expand_table 781 byug0cc5vn416 /* SQL Analyze(1) */ select /*+ full(t) no_parallel(t) no_parallel_index(t) 43 bkvvr4azs1n6z /* SQL Analyze(1) */ select /*+ full(t) no_parallel(t) no_parallel_index(t) 21 46sy4dfg3xbfn /* SQL Analyze(1) */ select /*+ full(t) no_parallel(t) no_parallel_index(t) 1559
So most queries are coming with “SQL Analyze” comment right in the beginning of SQL which is running from DBMS_STATS call, which is confusing. After some bug search I have found a MOS Doc ID 1480132.1 which includes a PL/SQL stack trace from a DBMS_STATS procedure call, and it is going up to DBMS_SQLTUNE_INTERNAL, which means DBMS_STATS has a call to the SQL Tuning package; very odd:
SQL> select * from dba_dependencies where name = 'DBMS_STATS_INTERNAL' and referenced_name = 'DBMS_SQLTUNE_INTERNAL'; OWNER NAME TYPE REFERENCED_OWNER REFERENCED_NAME ------------------------------ ------------------------------ ------------------ ------------------------------ ---------------------------------- REFERENCED_TYPE REFERENCED_LINK_NAME DEPE ------------------ ------------------------------------------------------------------------------------------------------------------------------- SYS DBMS_STATS_INTERNAL PACKAGE BODY SYS DBMS_SQLTUNE_INTERNAL PACKAGE HARD
Turns out, this call has nothing to do with the SQL Tuning. It is just a call to a procedure in this package, which happen to look like an SQL from SQL Tuning Advisor. I have traced a GATHER_TABLE_STATS call in a test database first with SQL trace and then with DBMS_HPROF and here is how the call tree looks like:
SELECT RPAD(' ', (level-1)*2, ' ') || fi.owner || '.' || fi.module AS name,
fi.function,
pci.subtree_elapsed_time,
pci.function_elapsed_time,
pci.calls
FROM dbmshp_parent_child_info pci
JOIN dbmshp_function_info fi ON pci.runid = fi.runid AND pci.childsymid = fi.symbolid
WHERE pci.runid = 1
CONNECT BY PRIOR childsymid = parentsymid
START WITH pci.parentsymid = 27;
NAME FUNCTION SUBTREE_ELAPSED_TIME FUNCTION_ELAPSED_TIME CALLS
---------------------------------------- ------------------------------ -------------------- --------------------- --------------------
...
SYS.DBMS_STATS_INTERNAL GATHER_SQL_STATS 21131962 13023 1
SYS.DBMS_ADVISOR __pkg_init 88 88 1
SYS.DBMS_SQLTUNE_INTERNAL GATHER_SQL_STATS 21118776 9440 1
SYS.DBMS_SQLTUNE_INTERNAL I_PROCESS_SQL 21107094 21104225 1
SYS.DBMS_LOB GETLENGTH 37 37 1
SYS.DBMS_LOB INSTR 42 42 1
SYS.DBMS_LOB __pkg_init 15 15 1
SYS.DBMS_SQLTUNE_INTERNAL I_VALIDATE_PROCESS_ACTION 74 39 1
SYS.DBMS_UTILITY COMMA_TO_TABLE 35 35 1
SYS.DBMS_SQLTUNE_UTIL0 SQLTEXT_TO_SIGNATURE 532 532 1
SYS.DBMS_SQLTUNE_UTIL0 SQLTEXT_TO_SQLID 351 351 1
SYS.XMLTYPE XMLTYPE 1818 1818 1
SYS.DBMS_SQLTUNE_UTIL0 SQLTEXT_TO_SQLID 528 528 1
SYS.DBMS_UTILITY COMMA_TO_TABLE 88 88 1
SYS.DBMS_UTILITY __pkg_init 10 10 1
SYS.SQLSET_ROW SQLSET_ROW 33 33 1
SYS.XMLTYPE XMLTYPE 1583 1583 1
SYS.DBMS_STATS_INTERNAL DUMP_PQ_SESSTAT 73 73 1
SYS.DBMS_STATS_INTERNAL DUMP_QUERY 2 2 1
...
So there is a procedure DBMS_SQLTUNE_INTERNAL.GATHER_SQL_STATS which is being called by DBMS_STATS_INTERNAL, and this procedure actually runs a SQL like this:
/* SQL Analyze(0) */ select /*+ full(t) no_parallel(t) no_parallel_index(t) dbms_stats cursor_sharing_exact use_weak_name_resl dynamic_sampling(0) no_monitoring no_substrb_pad */to_char(count("ID")),to_char(substrb(dump(min("ID"),16,0,32),1,120)),to_char(substrb(dump(max("ID"),16,0,32),1,120)),to_char(count("X")),to_char(substrb(dump(min("X"),16,0,32),1,120)),to_char(substrb(dump(max("X"),16,0,32),1,120)),to_char(count("Y")),to_char(substrb(dump(min("Y"),16,0,32),1,120)),to_char(substrb(dump(max("Y"),16,0,32),1,120)),to_char(count("PAD")),to_char(substrb(dump(min("PAD"),16,0,32),1,120)),to_char(substrb(dump(max("PAD"),16,0,32),1,120)) from "TIM"."T1" t /* NDV,NIL,NIL,NDV,NIL,NIL,NDV,NIL,NIL,NDV,NIL,NIL*/
Which is basically approximate NDV calculation. So, nothing to be afraid of, it’s just the way the code is organized: DBMS_STATS uses API of SQL Tuning framework when you are using DBMS_STATS.AUTO_SAMPLE_SIZE as the ESTIMATE_PERCENT (which is the default & recommended value in 11g+).