Oracle-活動會話數劇增

1. 現象

某天收到一數據庫活動會話數劇增不降，業務系統部分功能緩慢。

2. 分析

2.1 分析活動會話數變化趨勢

select 
 to_char(ash.sample_time, 'YYYY-MM-DD HH24:MI:SS') SAMPLE_TIME
,count(*) cnt
from dba_hist_active_sess_history ash
where ash.instance_number=1
 and ash.wait_class <> 'Idle'  /* 非空閑回話 **/
 and ash.sample_time between sysdate -1/2 and sysdate
 group by SAMPLE_TIME
 having count(*) >80 
 order by SAMPLE_TIME
;

2.2 抽一個時間點分析等待事件及SQL執行情況

select
   to_char(ash.sample_time, 'YYYY-MM-DD HH24:MI:SS') SAMPLE_TIME
   ,ash.sql_id
   ,ash.event
   ,count(*) CNT /* SQL的數量 **/
   ,TRUNC(SUM(TIME_WAITED) / 1000000,2) SECONDS_IN_WAIT /* SQL的等待時間 **/
   ,SUM(to_number(CAST(ash.sample_time AS DATE)-ash.sql_exec_start) *24*60*60 ) SECONDS_IN_EXECUTE /* SQL的執行時間 **/
from dba_hist_active_sess_history ash
where ash.instance_number=1
 and ash.wait_class <> 'Idle'
 and to_char(ash.sample_time,'YYYY-MM-DD HH24:MI:SS') = '2016-12-01 09:28:27'
group by
  ash.sample_time
 ,ash.sql_id
 ,ash.event
having count(*) > 5
order by SAMPLE_TIME,4
;

-- 發現174r8w7amsr17,83zyfck594vfk SQL執行時間較長，特別是174r8w7amsr17這條語句執行5+分鍾

2.3 分析語句執行過程

-- 83zyfck594vfk 語句執行情況
col event for a32
col p1text for a18
col p2text for a18
col p3text for a18

select
  to_char(ash.sample_time, 'YYYY-MM-DD HH24:MI:SS') SAMPLE_TIME
 ,ash.event
 ,ash.sql_id 
 ,ash.blocking_inst_id
 ,ash.blocking_session
 ,ash.blocking_session_serial#
 ,count(*) cnt
from dba_hist_active_sess_history ash 
where ash.instance_number=1
 and ash.wait_class <> 'idle'
 and ash.sql_id = '83zyfck594vfk'
 and to_char(ash.sample_time,'YYYY-MM-DD HH24:MI:SS') = '2016-12-01 09:28:27'
group by 
  sample_time
 ,ash.event
 ,ash.sql_id 
 ,ash.blocking_inst_id
 ,ash.blocking_session
 ,ash.blocking_session_serial#
order by sample_time,7
;

-- 發現session(1:4448,58047)阻塞了5個會話，下圖

2.4 分析下4448，58047會話的歷史執行情況

select 
 to_char(ash.sample_time, 'YYYY-MM-DD HH24:MI:SS') sample_time
,to_char(ash.sql_exec_start,'YYYY-MM-DD HH24:MI:SS') SQL_START_TIME
,ash.instance_number inst_id
,ash.session_id sid
,ash.session_serial# serial 
,ash.blocking_inst_id b_inst_id
,ash.blocking_session b_sid
,ash.blocking_session_serial# b_serial
,ash.sql_id
,ash.event
,to_number(CAST(ash.sample_time AS DATE)-ash.sql_exec_start) *24*60*60 SECONDS_IN_EXECUTE
,ash.xid
from dba_hist_active_sess_history ash
where 
 and ash.instance_number=1
 and ash.session_id = 4448
 and ash.session_serial# = 58047
order by sample_time
;

-- session(1:4448,58047) 從2016-12-01 09:25:33 建立了一個事務 (xid = '02FB001B0000619A') 持續到 2016-12-01 09:47:49 , 至少22分鍾里面執行 83zyfck594vfk 語句134次才提交,而每次執行時間挺小的，但整個事務執行時間過長，持續20多分鍾時間。而且83zyfck594vfk 語句執行存在“BY LOCAL INDEX ROWID”回表或是“INDEX SKIP SCAN ”操作，而索引跳躍掃描的性能不好。需要優化。

2.5 分析174r8w7amsr17語句是被前面這類多個83zyfck594vfk SQL長時間事務阻塞

3. 處理方案

3.1 優化83zyfck594vfk語句的索引；

3.2 檢查業務83zyfck594vfk功能模塊，分割大事務成多個小事務，減少阻塞。