[20191125]探究等待事件的本源.txt

[20191125]探究等待事件的本源.txt

–//当工作中遇到oracle的性能问题时,查看awr报表提供很好的解决问题途径.但是有时候很容易想当然.
–//比如以前我一看到 log file sync等待事件就很主观的认为redo 磁盘IO不行,实际上真实的情况可能有许多原因.
–//比如提交进程太多,cpu调度不过来.总之需要wait chains确定问题的本源.
–//上午看了大师Tanel Poder的dash_wait_chains.sql与ash_wait_chains.sql脚本,简单学习它的使用.
–//dash_wait_chains.sql 是基于DBA_HIST_ACTIVE_SESS_HISTORY视图.而ash_wait_chains.sql基于V$ACTIVE_SESSION_HISTORY视图.
–//二者命令执行格式差不多,仅仅学习ash_wait_chains.sql就ok了.

–//实际上根本不需要学习与记忆,看看脚本的开头就知道如何执行与使用:
— Usage:
—     @ash_wait_chains <grouping_cols> <filters> <fromtime> <totime>
—
— Example:
—     @ash_wait_chains username||”:”||program2||event2 session_type=”FOREGROUND” sysdate-1/24 sysdate
—
— Other:
—     This script uses only the in-memory V$ACTIVE_SESSION_HISTORY, use
—     @dash_wait_chains.sql for accessiong the DBA_HIST_ACTIVE_SESS_HISTORY archive
—
—     Oracle 10g does not  have the BLOCKING_INST_ID column in ASH so you”ll need
—     to comment out this column in this script. This may give you somewhat
—     incorrect results in RAC environment with global blockers.
—
——————————————————————————–

–//我做一点点简单的修改:
COL wait_chain FOR A300 WORD_WRAP
–//修改为:
COL wait_chain FOR A200 WORD_WRAP

–//我的工作环境最大显示宽度是271列.如果设置再小,字体显示就太小,我同事许多设置才237.
$ echo $COLUMNS
271

–//因为我还需要管理10g的数据库,根据前面的说明,可以拷贝脚本命名为ash_wait_chains10g.sql.
–//注解如下,当然对于rac环境可能不正确.
—         AND PRIOR d.blocking_inst_id = d.inst_id

–//作者还重新定义字段program,event为program2,event2.这样显示更加直观,占用宽度更小一些.
–//以下是测试部分:

1.环境:
qqqqqqqqqqq> @ ver1
PORT_STRING         VERSION    BANNER
——————- ———- —————————————————————————-
x86_64/Linux 2.4.xx 11.2.0.4.0 Oracle Database 11g Enterprise Edition Release 11.2.0.4.0 – 64bit Production

qqqqqqqqqqq> @ tpt/ash/ash_wait_chains  program2||”:”||event2 1=1 trunc(sysdate)+9/24 trunc(sysdate)+10/24
— Display ASH Wait Chain Signatures script v0.2 BETA by Tanel Poder ( http://blog.tanelpoder.com )
%This     SECONDS        AAS WAIT_CHAIN
—— ———- ———- ——————————————————————-
  33%        4362         .1 -> (XXXYYY.EXE) :ON CPU
  10%        1368          0 -> (NSSn) :LNS wait on SENDREQ
  10%        1323          0 -> (LGWR) :LGWR-LNS wait on channel
   6%         735          0 -> (XXXYYY.EXE) :log file sync  -> (LGWR) :LGWR-LNS wait on channel
   5%         605          0 -> (CAPAA-PIPE) :ON CPU
   5%         604          0 -> (XXXYYY.exe) :ON CPU
   4%         573          0 -> (wnwp.exe) :ON CPU
   3%         378          0 -> (DIAn) :ON CPU
   2%         274          0 -> (sqlplus) :ON CPU
   2%         246          0 -> (httpd.exe) :ON CPU
   2%         230          0 -> (PSPn) :ON CPU
   1%         115          0 -> (CKPT) :ON CPU
   1%         111          0 -> (LMSn) :ON CPU
   1%         111          0 -> (wnwp.exe) :log file sync  -> (LGWR) :LGWR-LNS wait on channel
   1%          89          0 -> (routine.exe) :ON CPU
   1%          87          0 -> (oracle) :ON CPU
   1%          86          0 -> (sqlplus) :control file sequential read
   1%          83          0 -> (LGWR) :ON CPU
   1%          76          0 -> (routine.exe) :log file sync  -> (LGWR) :LGWR-LNS wait on channel
   1%          72          0 -> (ARCn) :ON CPU
   1%          69          0 -> (XXXYYY.EXE) :gc current block 2-way
   1%          68          0 -> (XXXYYY.EXE) :gc cr block 2-way
   0%          63          0 -> (LGWR) :log file parallel write
   0%          55          0 -> (LMON) :ON CPU
   0%          55          0 -> () :null event
   0%          47          0 -> (XXXYYY.EXE) :log file sync
   0%          46          0 -> (nnnn.exe) :log file sync  -> (LGWR) :LGWR-LNS wait on channel
   0%          39          0 -> (LMSn) :gcs log flush sync  -> (LGWR) :LGWR-LNS wait on channel
   0%          38          0 -> (XXXYYY.EXE) :direct path read
   0%          38          0 -> (XXXYYY.EXE) :gc cr block busy
30 rows selected.

–//开始不理解AAS表示什么.表示如下
ROUND(COUNT(*) / ((CAST(&4 AS DATE) – CAST(&3 AS DATE)) * 86400), 1) AAS
–//相当于每秒取样几次.
–//参数3,参数4是时间范围,我建议使用类似例子的格式.你仅仅需要知道1/24表示1小时,1/1440表示1分钟就ok了.
–//比如你需要查看当天8:15到9:10的情况,时间可以写成:
trunc(sysdate)+8/24+15/1440 trunc(sysdate)+9/24+10/1440
–//另外注意一点:V$ACTIVE_SESSION_HISTORY保留信息有限制,时间太久的可能查询不到.要改用dash_wait_chains.sql脚本查询.
–//参数2可以加入一些过滤例子:session_type=”FOREGROUND” 或者使用 1=1 表示全部.

–//你可以看到这台服务器的log file sync等待事件实际上由于(LGWR) :LGWR-LNS wait on channel.
–//问题在与安装配置dg时的参数:

qqqqqqqqqqq> show parameter log_archive_dest_2
NAME               TYPE   VALUE
—————— —— —————————————————————————————————-
log_archive_dest_2 string service=rzdbra lgwr sync reopen=15 max_failure=10 net_timeout=30 optional noaffirm db_unique_name=XXXXXX
–//对方配置了sync,这样在提交时要确定dg方已经写入日志文件才算提交.
–//如果改成async,相关等待LNS wait on SENDREQ,LGWR-LNS wait on channel,log file sync都会消失或者减少.

–//再来看看另外1个10g的数据库:
WWWWWWWWW> @ ver1
PORT_STRING         VERSION        BANNER
——————- ————– —————————————————————-
x86_64/Linux 2.4.xx 10.2.0.4.0     Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 – 64bi

WWWWWWWWW> @ tpt/ash/ash_wait_chains10g  program2||”:”||event2 1=1 trunc(sysdate)+9/24 trunc(sysdate)+10/24
— Display ASH Wait Chain Signatures script v0.2 BETA by Tanel Poder ( http://blog.tanelpoder.com )
%This     SECONDS        AAS WAIT_CHAIN
—— ———- ———- ————————————————————————
  30%        3602         .1 -> (xxxzzzz.exe) :db file sequential read
  19%        2244          0 -> (DBWn) :ON CPU
  11%        1321          0 -> (LGWR) :log file parallel write
   9%        1113          0 -> (xxxzzzz.exe) :ON CPU
   7%         890          0 -> (xxxzzzz.exe) :db file scattered read
   6%         784          0 -> (xxxzzzz.exe) :log file sync  -> (LGWR) :log file parallel write
   5%         633          0 -> (aaatobbbb.exe) :log file sync  -> (LGWR) :log file parallel write
   4%         490          0 -> (CKPT) :control file parallel write
   1%         106          0 -> (wnwp.exe) :db file sequential read
   1%          88          0 -> (JDBC Thin Client) :db file sequential read
   1%          86          0 -> (aaatobbbb.exe) :db file scattered read
   1%          82          0 -> (mnnn) :db file sequential read
   0%          59          0 -> (xxxzzzz.exe) :read by other session
   0%          55          0 -> (wnwp.exe) :ON CPU
   0%          54          0 -> (JDBC Thin Client) :ON CPU
   0%          45          0 -> (plsqldev.exe) :enq: TX – row lock contention
   0%          44          0 -> (sqlplus) :ON CPU
   0%          27          0 -> (wnwp.exe) :log file sync  -> (LGWR) :log file parallel write
   0%          27          0 -> (aaatobbbb.exe) :ON CPU
   0%          25          0 -> (xxxzzzz.exe) :SQL*Net more data from dblink
   0%          25          0 -> (mnnn) :ON CPU
   0%          24          0 -> (ARCn) :log file sequential read
   0%          19          0 -> (xxxzzzz.exe) :db file parallel read
   0%          13          0 -> (CTWR) :ON CPU
   0%          13          0 -> (aaatobbbb.exe) :db file sequential read
   0%          12          0 -> (Jnnn) :db file sequential read
   0%          10          0 -> (aaatobbbb.exe) :log file sync
   0%          10          0 -> (SMON) :db file sequential read
   0%          10          0 -> (ARCn) :ON CPU
   0%          10          0 -> (CJQn) :ON CPU
30 rows selected.
–//可以发现这台系统磁盘io已经不行.如果负荷在增加就不行.顺便说一下这台机器硬件配置很差.

WWWWWWWWW> @ tpt/ash/ash_wait_chains10g  event2 1=1 trunc(sysdate)+7/24 sysdate
— Display ASH Wait Chain Signatures script v0.2 BETA by Tanel Poder ( http://blog.tanelpoder.com )
%This     SECONDS        AAS WAIT_CHAIN
—— ———- ———- ———————————————————–
  32%        4910         .1 -> db file sequential read
  32%        4887         .1 -> ON CPU
  12%        1896          0 -> log file sync  -> log file parallel write
  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  11%        1761          0 -> log file parallel write
   7%        1078          0 -> db file scattered read
   4%         650          0 -> control file parallel write
   0%          59          0 -> read by other session
   0%          45          0 -> enq: TX – row lock contention
   0%          34          0 -> SQL*Net more data to client
   0%          34          0 -> SQL*Net more data from dblink
   0%          24          0 -> log file sequential read
   0%          23          0 -> db file parallel read
   0%          21          0 -> log file sync
   0%           7          0 -> log file switch completion
   0%           6          0 -> null event
   0%           5          0 -> change tracking file synchronous write
   0%           4          0 -> SQL*Net more data from client
   0%           4          0 -> control file sequential read
   0%           3          0 -> os thread startup
   0%           2          0 -> enq: CF – contention  -> control file parallel write
   0%           1          0 -> SQL*Net break/reset to client
   0%           1          0 -> Log archive I/O
   0%           1          0 -> latch: library cache
   0%           1          0 -> log file sync  -> ON CPU
24 rows selected.

–//注意看下划线,log file sync主要是由于log file parallel write太慢造成的,与前面的不同.这样定位问题就不会太盲目乱猜.