Oracle導致Redo日志暴增的SQL語句排查

一、概述

最近數據庫頻繁不定時的報出一些耗時長的SQL，甚至SQL執行時間過長，導致連接斷開現象。下面是一些排查思路。

二、查詢日志的大小，日志組情況

SELECT L.GROUP#,
       LF.MEMBER,
       L.ARCHIVED,
       L.BYTES / 1024 / 1024 "SIZE(M)",
       L.MEMBERS
FROM V$LOG L,
     V$LOGFILE LF
WHERE L.GROUP# = LF.GROUP#;

查詢結果：

 

 從上圖可以看出目前共分為10個日志組，每個日志組2個文件，每個文件大小為3G。

三、查詢Oracle最近幾天每小時歸檔日志產生數量

SELECT SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH:MI:SS'), 1, 5)                             Day,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '00', 1, 0)) H00,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '01', 1, 0)) H01,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '02', 1, 0)) H02,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '03', 1, 0)) H03,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '04', 1, 0)) H04,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '05', 1, 0)) H05,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '06', 1, 0)) H06,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '07', 1, 0)) H07,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '08', 1, 0)) H08,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '09', 1, 0)) H09,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '10', 1, 0)) H10,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '11', 1, 0)) H11,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '12', 1, 0)) H12,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '13', 1, 0)) H13,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '14', 1, 0)) H14,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '15', 1, 0)) H15,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '16', 1, 0)) H16,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '17', 1, 0)) H17,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '18', 1, 0)) H18,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '19', 1, 0)) H19,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '20', 1, 0)) H20,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '21', 1, 0)) H21,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '22', 1, 0)) H22,
       SUM(DECODE(SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH24:MI:SS'), 10, 2), '23', 1, 0)) H23,
       COUNT(*)                                                                           TOTAL
FROM v$log_history a
WHERE first_time >= to_char(sysdate - 10)
GROUP BY SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH:MI:SS'), 1, 5)
ORDER BY SUBSTR(TO_CHAR(first_time, 'MM/DD/RR HH:MI:SS'), 1, 5) DESC;

查詢結果

 

從上圖可以看出業務高峰期每小時產生40個日志文件左右（目前設定的每個日志文件大小為3G），平均1.5分鍾產生一個3G的日志文件。而oracle官方建議Redo日志平均30分鍾切換一次最好。

四、查看最近2小時"塊改變"最多的segment

redo大量產生必然是由於大量產生"塊改變"。從awr視圖中找到"塊改變"最多的segment。

這是查詢最近2小時(120分鍾)的，begin_interval_time> sysdate - 120/1440，大家也可以自定義修改查詢最近多少分鍾的。

select *
from (SELECT to_char(begin_interval_time, 'YYYY_MM_DD HH24:MI') snap_time,
             dhsso.object_name,
             SUM(db_block_changes_delta)
      FROM dba_hist_seg_stat dhss,
           dba_hist_seg_stat_obj dhsso,
           dba_hist_snapshot dhs
      WHERE dhs.snap_id = dhss.snap_id
        AND dhs.instance_number = dhss.instance_number
        AND dhss.obj# = dhsso.obj#
        AND dhss.dataobj# = dhsso.dataobj#
        AND begin_interval_time > sysdate - 120 / 1440
      GROUP BY to_char(begin_interval_time, 'YYYY_MM_DD HH24:MI'), dhsso.object_name
      order by 3 desc)
where rownum <= 5;

查詢結果：

 

五、從awr視圖中找出步驟四中排序靠前的對象涉及的SQL

說明：LIKE '%MON_MODS$%'中MON_MODS是步驟1中查詢出來的OBJECT_NAME

SELECT to_char(begin_interval_time, 'YYYY_MM_DD HH24:MI'),
       dbms_lob.substr(sql_text, 4000, 1),
       dhss.instance_number,
       dhss.sql_id,
       executions_delta,
       rows_processed_delta
FROM dba_hist_sqlstat dhss,
     dba_hist_snapshot dhs,
     dba_hist_sqltext dhst
WHERE UPPER(dhst.sql_text) LIKE '%MON_MODS$%'
  AND dhss.snap_id = dhs.snap_id
  AND dhss.instance_Number = dhs.instance_number
  AND dhss.sql_id = dhst.sql_id;

查詢結果

 

六、從ASH相關視圖找到執行這些SQL的session、module和machine

select * from dba_hist_active_sess_history WHERE sql_id = 'c9n8kv7afchtd';
select * from v$active_session_history where sql_Id = 'c9n8kv7afchtd';

c9n8kv7afchtd是SQL_ID,替換第二步查詢的結果SQL_ID列

 七、排查問題SQL

通過第四步，我們確定了導致產生大量redo日志主要涉及三張表，再通過第五步確定了每張表排名前五的SQL。針對這些產生大量Redo日志的SQL，就是需要做優化的地方。