數據庫Oracle 11.2.0.4 RAC 2節點，業務反饋SQL執行緩慢，發現執行計划是全表掃描，前一天是走索引很快，猜測是統計信息不准確導致。
后續對分區表99G，收集統計信息后，業務反饋SQL走索引，恢復正常。
一、實際上收集統計信息的操作
觀察如下鏈接
http://www.oracleplus.net/arch/1158.html
https://www.cnblogs.com/kawashibara/p/9762724.html
小結
收集統計信息，使用cascade=>true，不加分區參數，會對表全局，及級聯的索引全局索引收集統計信息；
partname=>'PARTTABLE_1998'，只對指定分區的統計信息進行改變，表的全局，及其它分區信息不改變；
granularity=>'GLOBAL'，只修改分區表的全局統計信息，但是具體的分區統計信息不改變；
granularity=>'AUTO'，默認,oracle自動根據分區類型選擇粒度，會對全局和所有分區統計信息均發生改變；
granularity=>'ALL'，分區表全局+分區表均收集統計信息，明確指定；
SQL> select owner,table_name,PARTITIONED,last_analyzed from dba_tables where owner='Oxx' ORDER BY 3; OWNER TABLE_NAME PAR LAST_ANALYZED ------------------------------ ------------------------------ --- ------------------- Oxx OMxx NO 2020-05-12 22:00:06 Oxx OMxx_LOG YES 2020-05-13 13:59:33 [xx]$ cat analy.sh sqlplus / as sysdba <<EOF set timing on set time on spool /home/oracle/analyze.log select sysdate from dual; exec DBMS_STATS.GATHER_TABLE_STATS(ownname=>'xx', tabname=>'xx',granularity=>'ALL',cascade=>true,estimate_percent=>30,degree=>6); exec DBMS_STATS.GATHER_TABLE_STATS(ownname=>'xx', tabname=>'xx',cascade=>true,estimate_percent=>30,degree=>6); exit; EOF -- Elapsed: 00:52:19.97
Elapsed: 00:02:13.45
--查詢段大小
select sum(bytes)/1024/1024/1024 from dba_segments where owner='xx' and segment_name='xx';
--查詢session信息
set pagesize 2000
set linesize 210
col sql_id for a14
col s_time for a12
col status for a6
col event format a30
col username for a14
col osuser for a8
col p1 for 9999999
col sid for 9999
col p2 for 999999
col p3 for 9999999999999
col program format a25
Col machine for a20
col serial# format 99999
select substr(b.program, 1, 25) program, b.sid, b.serial#, b.username, substr(b.osuser, 1, 8) osuser,
b.sql_id, substr(b.machine, 1, 20) machine, b.status, to_char(b.SQL_EXEC_START, 'dd hh24:mi:ss') s_time,
(case when b.STATE = 'WAITING' then b.event else 'CPU' end) event from v$session_wait a, v$session b
where a.sid = b.sid and status = 'ACTIVE' and not (b.type = 'BACKGROUND' and b.state = 'WAITING'
and b.wait_class = 'Idle') Order by program, sid, sql_id;
--查詢統計信息收集進度
set linesize 200 pagesize 50
col opname for a40
col target for a30
col message for a40
select sid,opname,target,sofar,totalwork,round(sofar/totalwork,4)*100 per,message FROM
V$SESSION_LONGOPS where (sid,serial#) in (select sid,serial# from v$session where type='USER')
and sofar<>totalwork order by sid,serial#;
二、分區表統計信息收集測試
收集統計信息的參數變化可以看如下鏈接 https://www.cnblogs.com/jerome-lamb/p/7535014.html --創建測試對象 create table range_part_tab(id number, deal_date date, area_code number, contents varchar2(4000)) partition by range(deal_date) ( partition p1 values less than(to_date('2017-02-01','yyyy-mm-dd')), partition p2 values less than(to_date('2017-03-01','yyyy-mm-dd')), partition p3 values less than(to_date('2017-04-01','yyyy-mm-dd')), partition p4 values less than(to_date('2017-05-01','yyyy-mm-dd')), partition p5 values less than(to_date('2017-06-01','yyyy-mm-dd')), partition p6 values less than(to_date('2017-07-01','yyyy-mm-dd')), partition p7 values less than(to_date('2017-08-01','yyyy-mm-dd')), partition p8 values less than(to_date('2017-09-01','yyyy-mm-dd')), partition p9 values less than(to_date('2017-10-01','yyyy-mm-dd')), partition p10 values less than(to_date('2017-11-01','yyyy-mm-dd')), partition p11 values less than(to_date('2017-12-01','yyyy-mm-dd')), partition p12 values less than(to_date('2018-01-01','yyyy-mm-dd')), partition p_max values less than(maxvalue)) ; insert into range_part_tab (id,deal_date,area_code,contents) select rownum, to_date(to_char(sysdate-365,'J')+ trunc(dbms_random.value(0,365)),'J'), ceil(dbms_random.value(590,599)), rpad('*',400,'*') from dual connect by rownum <= 100000; SQL> r 1* insert into RANGE_PART_TAB select * from RANGE_PART_TAB commit; SQL> select sum(bytes)/1024/1024 from user_segments; SUM(BYTES)/1024/1024 -------------------- 1539 SQL> create index RANGE_PART_TAB_ind_id on RANGE_PART_TAB(id) local; SQL> select TABLE_NAME,STATUS,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,PARTITIONED,LAST_ANALYZED from user_tables; TABLE_NAME STATUS NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE PAR LAST_ANALYZED ------------------------------ -------- ---------- ---------- ------------ ----------- ------------- --- RANGE_PART_TAB VALID YES --創建索引后,會自動收集索引的統計信息，但是不包括表的！ SQL> select index_name,TABLE_NAME,BLEVEL,LEAF_BLOCKS,DISTINCT_KEYS,STATUS,NUM_ROWS,SAMPLE_SIZE,LAST_ANALYZED,PARTITIONED from user_indexes; INDEX_NAME TABLE_NAME BLEVEL LEAF_BLOCKS DISTINCT_KEYS STATUS NUM_ROWS SAMPLE_SIZE LAST_ANAL PAR -------------------------------------------------------------------------------- -------- ------------------------------ RANGE_PART_TAB_IND_ID RANGE_PART_TAB 2 7094 100000 N/A 3200000 3200000 19-MAY-20 YES SQL> select TABLE_NAME,PARTITION_NAME,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,LAST_ANALYZED,HIGH_VALUE from user_tab_partitions order by 2; TABLE_NAME PARTITION_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE LAST_ANAL HIGH_VALUE ------------------------------ ------------------------------ ---------- ---------- ------------ ----------- --------- -------------------------------------------------------------------------------- RANGE_PART_TAB P1 TO_DATE(' 2017-02-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P10 TO_DATE(' 2017-11-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P11 TO_DATE(' 2017-12-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P12 TO_DATE(' 2018-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P2 TO_DATE(' 2017-03-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P3 TO_DATE(' 2017-04-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P4 TO_DATE(' 2017-05-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P5 TO_DATE(' 2017-06-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P6 TO_DATE(' 2017-07-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P7 TO_DATE(' 2017-08-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P8 TO_DATE(' 2017-09-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P9 TO_DATE(' 2017-10-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P_MAX MAXVALUE 13 rows selected. SQL> select index_name,PARTITION_NAME,STATUS,BLEVEL,LEAF_BLOCKS,DISTINCT_KEYS,NUM_ROWS,SAMPLE_SIZE,LAST_ANALYZED from user_ind_partitions; INDEX_NAME PARTITION_NAME STATUS BLEVEL LEAF_BLOCKS DISTINCT_KEYS NUM_ROWS SAMPLE_SIZE LAST_ANAL ------------------------------ ------------------------------ -------- ---------- ----------- ------------- ---------- ----------- --------- RANGE_PART_TAB_IND_ID P_MAX USABLE 2 7094 100000 3200000 3200000 19-MAY-20 RANGE_PART_TAB_IND_ID P9 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P8 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P7 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P6 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P5 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P4 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P3 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P2 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P12 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P11 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P10 USABLE 0 0 0 0 19-MAY-20 RANGE_PART_TAB_IND_ID P1 USABLE 0 0 0 0 19-MAY-20 13 rows selected. 收集統計信息 exec DBMS_STATS.GATHER_TABLE_STATS(ownname=>'YZ', tabname=>'RANGE_PART_TAB',cascade=>true,estimate_percent=>30,degree=>6); SQL> select TABLE_NAME,PARTITION_NAME,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,LAST_ANALYZED,HIGH_VALUE from user_tab_partitions order by 2; TABLE_NAME PARTITION_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE LAST_ANAL HIGH_VALUE ------------------------------ ------------------------------ ---------- ---------- ------------ ----------- --------- -------------------------------------------------------------------------------- RANGE_PART_TAB P1 0 0 0 19-MAY-20 TO_DATE(' 2017-02-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P10 0 0 0 19-MAY-20 TO_DATE(' 2017-11-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P11 0 0 0 19-MAY-20 TO_DATE(' 2017-12-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P12 0 0 0 19-MAY-20 TO_DATE(' 2018-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P2 0 0 0 19-MAY-20 TO_DATE(' 2017-03-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P3 0 0 0 19-MAY-20 TO_DATE(' 2017-04-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P4 0 0 0 19-MAY-20 TO_DATE(' 2017-05-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P5 0 0 0 19-MAY-20 TO_DATE(' 2017-06-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P6 0 0 0 19-MAY-20 TO_DATE(' 2017-07-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P7 0 0 0 19-MAY-20 TO_DATE(' 2017-08-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P8 0 0 0 19-MAY-20 TO_DATE(' 2017-09-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P9 0 0 0 19-MAY-20 TO_DATE(' 2017-10-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P_MAX 3198140 196328 0 959442 19-MAY-20 MAXVALUE 13 rows selected. SQL> select TABLE_NAME,STATUS,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,PARTITIONED,LAST_ANALYZED from user_tables; TABLE_NAME STATUS NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE PAR LAST_ANAL ------------------------------ -------- ---------- ---------- ------------ ----------- --- --------- RANGE_PART_TAB VALID 3195423 196328 0 958627 YES 19-MAY-20 收集統計信息后,可以發現表分區的統計信息及表的總體統計信息都有了。 2.數據變更 insert into range_part_tab (id,deal_date,area_code,contents) select rownum, to_date(to_char(to_date('2018-03-01','yyyy-mm-dd')-365,'J')+ trunc(dbms_random.value(0,365)),'J'), ceil(dbms_random.value(590,599)), rpad('*',400,'*') from dual connect by rownum <= 100000; SQL> commit; SQL> select count(*) from RANGE_PART_TAB; COUNT(*) ---------- 3300000 10萬新增數據,總大小是330萬,3%的數據變更，達不到統計信息過舊的條件。 SQL> select table_name,PARTITION_NAME,STALE_STATS from user_TAB_STATISTICS; TABLE_NAME PARTITION_NAME STA ------------------------------ ------------------------------ --- RANGE_PART_TAB NO RANGE_PART_TAB P1 NO RANGE_PART_TAB P3 NO RANGE_PART_TAB P8 NO RANGE_PART_TAB P12 NO RANGE_PART_TAB P2 NO RANGE_PART_TAB P7 NO RANGE_PART_TAB P6 NO RANGE_PART_TAB P9 NO RANGE_PART_TAB P5 NO RANGE_PART_TAB P11 NO RANGE_PART_TAB P10 NO RANGE_PART_TAB P4 NO RANGE_PART_TAB P_MAX NO 14 rows selected. 收集10%粒度！ exec DBMS_STATS.GATHER_TABLE_STATS(ownname=>'YZ', tabname=>'RANGE_PART_TAB',cascade=>true,estimate_percent=>10,degree=>6); SQL> select TABLE_NAME,PARTITION_NAME,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,LAST_ANALYZED,HIGH_VALUE from user_tab_partitions order by 2; TABLE_NAME PARTITION_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE LAST_ANAL HIGH_VALUE ------------------------------ ------------------------------ ---------- ---------- ------------ ----------- --------- -------------------------------------------------------------------------------- RANGE_PART_TAB P1 0 0 0 19-MAY-20 TO_DATE(' 2017-02-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P10 8441 558 0 5079 19-MAY-20 TO_DATE(' 2017-11-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P11 8153 494 0 4836 19-MAY-20 TO_DATE(' 2017-12-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P12 8443 558 0 5161 19-MAY-20 TO_DATE(' 2018-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P2 0 0 0 19-MAY-20 TO_DATE(' 2017-03-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P3 8472 558 0 5061 19-MAY-20 TO_DATE(' 2017-04-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P4 8261 494 0 5316 19-MAY-20 TO_DATE(' 2017-05-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P5 8511 558 0 5260 19-MAY-20 TO_DATE(' 2017-06-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P6 8290 494 0 5049 19-MAY-20 TO_DATE(' 2017-07-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P7 8382 558 0 4919 19-MAY-20 TO_DATE(' 2017-08-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P8 8538 494 0 4993 19-MAY-20 TO_DATE(' 2017-09-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P9 8266 494 0 4868 19-MAY-20 TO_DATE(' 2017-10-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P_MAX 3211810 196328 0 321181 19-MAY-20 MAXVALUE 13 rows selected. 加上分區級聯收集參數？ exec DBMS_STATS.GATHER_TABLE_STATS(ownname=>'YZ', tabname=>'RANGE_PART_TAB',granularity=>'ALL',cascade=>true,estimate_percent=>30,degree=>6); SQL> select TABLE_NAME,PARTITION_NAME,NUM_ROWS,BLOCKS,EMPTY_BLOCKS,SAMPLE_SIZE,LAST_ANALYZED,HIGH_VALUE from user_tab_partitions order by 2; TABLE_NAME PARTITION_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS SAMPLE_SIZE LAST_ANAL HIGH_VALUE ------------------------------ ------------------------------ ---------- ---------- ------------ ----------- --------- -------------------------------------------------------------------------------- RANGE_PART_TAB P1 0 0 0 19-MAY-20 TO_DATE(' 2017-02-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P10 8303 558 0 5010 19-MAY-20 TO_DATE(' 2017-11-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P11 8344 494 0 5100 19-MAY-20 TO_DATE(' 2017-12-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P12 8620 558 0 5218 19-MAY-20 TO_DATE(' 2018-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P2 0 0 0 19-MAY-20 TO_DATE(' 2017-03-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P3 8278 558 0 4973 19-MAY-20 TO_DATE(' 2017-04-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P4 8189 494 0 5024 19-MAY-20 TO_DATE(' 2017-05-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P5 8530 558 0 2559 19-MAY-20 TO_DATE(' 2017-06-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P6 8161 494 0 4976 19-MAY-20 TO_DATE(' 2017-07-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P7 8793 558 0 2638 19-MAY-20 TO_DATE(' 2017-08-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P8 8520 494 0 2556 19-MAY-20 TO_DATE(' 2017-09-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P9 8483 494 0 2545 19-MAY-20 TO_DATE(' 2017-10-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIA RANGE_PART_TAB P_MAX 3217647 196328 0 965294 19-MAY-20 MAXVALUE 13 rows selected. 有一些誤差，但是影響不是很大？在數據不均勻的情況下,使用Oracle自動收集粒度,易造成采樣比例過少,最終導致數據變化過大。 <1g 100% 1g~5g 30% >5g 10%
根據實際測試，可以發現使用auto參數就是收集分區表統計信息，不加granularity=>'ALL'明確指定，則默認使用AUTO會自動對表全局、表分區均收集統計信息，但是粒度由Oracle自行控制,如果使用ALL后就是明確指定表對象、表分區均百分百收集，具體比例看參數控制。
在生產環境100G大表的情況下，使用AUTO默認情況下，很可能造成數據收集的信息不准確，最終導致影響SQL的執行，建議使用granularity=>'ALL’明確