ogg是逻辑同步，不想ADG利用后镜像直接修改block中的内容，ogg是根据redo片中SQL，以及SQL对应的后镜像值进行表的修改，这样如果大表上面没有索引，或者走错索引就会导致同步慢。但是OGG都是单表update，所以很容易优化SQL。以下利用sql profile固定执行计划，清除sharedpool对应的执行计划达到重新利用正常的执行计划
-bash-3.2$ ogg
Oracle GoldenGate Command Interpreter for Oracle
Version 11.2.1.0.6 16211226 OGGCORE_11.2.1.0.6_PLATFORMS_130418.1829_FBO
Solaris, sparc, 64bit (optimized), Oracle 11g on Apr 22 2013 15:23:39
Copyright (C) 1995, 2013, Oracle and/or its affiliates. All rights reserved.
GGSCI (ODSDB) 1> info all
Program Status Group Lag at Chkpt Time Since Chkpt
MANAGER RUNNING
JAGENT STOPPED
REPLICAT RUNNING RP10 00:00:00 00:00:03
REPLICAT RUNNING RP7 00:00:06 00:00:03
REPLICAT RUNNING RPS1 42:41:14 00:00:01
REPLICAT RUNNING RPS2 00:00:00 00:00:10
REPLICAT RUNNING RPS2A 00:00:00 00:00:00
REPLICAT RUNNING RPS2B 00:00:00 00:00:00
REPLICAT RUNNING RPS3 00:00:00 00:00:02
REPLICAT RUNNING RPS3A 00:00:00 00:00:03
REPLICAT RUNNING RPS3B 00:00:00 00:00:02
REPLICAT RUNNING RPS4 00:00:00 00:00:07
REPLICAT RUNNING RPS4A 00:00:00 00:00:07
REPLICAT RUNNING RPS4B 00:00:00 00:00:07
REPLICAT RUNNING RPS4C 00:00:00 00:00:07
REPLICAT RUNNING RPS4D 00:00:00 00:00:08
REPLICAT RUNNING RPS4E 00:00:00 00:00:07
REPLICAT RUNNING RPS4F 00:00:00 00:00:07
REPLICAT RUNNING RPS4G 00:00:00 00:00:07
REPLICAT RUNNING RPS4H 00:00:00 00:00:07
REPLICAT RUNNING RPS4I 00:00:00 00:00:08
REPLICAT RUNNING RPS4J 00:00:00 00:00:07
跟他去了终端旁边，发现确实有个rps1进程延迟有点高。看着进程状态是RUNNING，但是根据我维护六年ogg来看，有可能是假象，那么怎么判断呢？
1，查看ggserr.log
2，查看dirrpt下面该进程的dsc文件
3，info 进程名多次，看看RBA是否有变化
经过以上几步rps1进程状态是正常的，延迟高问题出在哪里？view params rps1看到该进程只有一张表。那我们看看这个进程到底卡在哪里了
SQL> !ps -ef |grep rps1 |grep -v grep
orao 12169 11012 0 Jun 07 ? 614:58 /odogg/baseogg/replicat PARAMFILE /odsogg/baseogg/dirprm/rps1.prm REPORTFILE /
SQL> !ps -ef |grep 12169 |grep -v grep
orao 12170 12169 1 Jun 07 ? 2315:30 oracleodb (DESCRIPTION=(LOCAL=YES)(ADDRESS=(PROTOCOL=beq)))
orao 12169 11012 0 Jun 07 ? 614:58 /odogg/baseogg/replicat PARAMFILE /odsogg/baseogg/dirprm/rps1.prm REPORTFILE /
SQL> select s.sid,s.SERIAL#,sql_id from v$process p,v$session s where p.addr=s.paddr and p.spid=12170;
SID SERIAL# SQL_ID
---------- ---------- -------------
5146 113 5099dwmx3s4mf
SQL> /
SID SERIAL# SQL_ID
---------- ---------- -------------
5146 113 dzubcf8jm69yx
SQL> /
SID SERIAL# SQL_ID
---------- ---------- -------------
5146 113 dzubcf8jm69yx
SQL> /
SID SERIAL# SQL_ID
---------- ---------- -------------
5146 113 dzubcf8jm69yx
询问了一下维护过该系统的同事，说是不是统计信息太旧。顺着思路往下走走
SQL> select OWNER,TABLE_NAME,LAST_ANALYZED from dba_tables where table_name='TF_F_USER_REVGRPTAG';
OWNER TABLE_NAME LAST_ANALYZED
-------------------------------------------------------------------------------
TTT TF_F_USER_REXXX 2017-06-26 00:20:56
SQL> select table_name,num_rows from dba_tables where owner='TTT' and table_name='TF_F_USER_REXXX';
TABLE_NAME NUM_ROWS
------------------------------ ----------
TF_F_USER_REXXX 409269832
SQL> select count(*) from TTT.TF_F_USER_REXXX;
COUNT(*)
----------
409955510
接下来我们看看该sql执行计划是不是走错了？
两条执行计划，很明显137632316运行了5400次但是执行时间却远远高于第二个执行计划。
用sql profile绑定一下执行计划
SQL> @coe_xfr_sql_profile.sql
Parameter 1:
SQL_ID (required)
Enter value for 1: dzubcf8jm69yx
PLAN_HASH_VALUE AVG_ET_SECS
--------------- -----------
2320424968 .004
137632316 21.119
Parameter 2:
PLAN_HASH_VALUE (required)
Enter value for 2: 2320424968
Values passed to coe_xfr_sql_profile:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
SQL_ID : "dzubcf8jm69yx"
PLAN_HASH_VALUE: "2320424968"
查看是否绑定成功
SQL>select name from dba_sql_profiles;
NAME
------------------------------
coe_dzubcf8jm69yx_2320424968
SYS_SQLPROF_015d075ccdd40000
coe_a5hgtqfq09tcu_229988255
在将内存中的该sql语句清除出去
SQL>select address,hash_value,executions,parse_calls from v$sql where sql_TEXT like 'UPDATE "TTT"."TF_F_USER_RExxxx" SET "PARTITION_ID" = :a27,"USER_ID" = :a28%';
ADDRESS HASH_VALUE EXECUTIONS PARSE_CALLS
---------------- ---------- ---------- -----------
00000017673EA5E8 4198240878 100 0
00000017CF3E0000 897328139 673 0
00000017982D40D0 1588244011 8 0
000000049D04B1A8 590555101 10 0
00000017AA525230 4271852353 2 0
0000001777FC6A20 3666847391 735 0
00000017B9154A58 1180290247 1 0
7 rows selected.
SQL>alter session set events '5614566 trace name context forever';
Session altered.
SQL>exec dbms_shared_pool.purge('00000017673EA5E8,4198240878','C');
PL/SQL procedure successfully completed.
SQL>exec dbms_shared_pool.purge('00000017CF3E0000,897328139','C');
exec db
PL/SQL procedure successfully completed.
SQL>ms_shared_pool.purge('00000017982D40D0,1588244011','C');
PL/SQL procedure successfully completed.
SQL>exec dbms_shared_pool.purge('000000049D04B1A8,590555101','C');
PL/SQL procedure successfully completed.
SQL>exec dbms_shared_pool.purge('00000017AA525230,4271852353','C');
PL/SQL procedure successfully completed.
SQL>exec dbms_shared_pool.purge('0000001777FC6A20,3666847391','C');
PL/SQL procedure successfully completed.
SQL>exec dbms_shared_pool.purge('00000017B9154A58,1180290247','C');
PL/SQL procedure successfully completed.
--------
清理过程还可以通过DDL表改动做到强制硬解析
细粒度对表DDL操作
–收集统计信息后,OGG复制执行的SQL执行计划没有改变,对表DDL,让SQL硬解析
comment on table xx.xx is 'ogg_big_tab';
select owner,to_char(LAST_DDL_TIME,'yyyy-mm-dd hh24:mi:ss')
from dba_objects where object_name='xx';
------------------------------------------------------------
xx 2021-11-14 02:54:45
由于只是修改表的注释,所以实际操作是毫秒级别,不会像其它DDL操作执行几分钟、甚至几个小时阻塞业务正常DML操作。
-----------
keycols(primary_column_name); 作用是强制根据这个列做改动？
1.表存在主键 select * from user_cons_columns
where constraint_name = (select constraint_name from user_constraints
where table_name = 'BST_FAVORITE' and constraint_type ='P'); 2.对OGG复制进程添加参数指定主键列 map source_owner.table_name target target_owner.table_name; 添加参数
map source_owner.table_name target target_owner.table_name，keycols(primary_column_name);
3.对于不存在主键的表呢？？？ select count(*) from xxx; 得到表的数量，如果表很大，不执行最好。 通过dba_tab_columns 根据NUM_DISTINCT 得到最多distinct的列，及选择性好的列。
SQL> select COLUMN_NAME,NUM_NULLS,NUM_DISTINCT,to_char(LAST_ANALYZED,'yyyy-mm-dd') as "date" from dba_tab_columns where owner='cc' and table_name='cc' order by 3; 结合表的索引列selecta.uniqueness 索引类型,b.index_name 索引名称,b.column_name 字段fromuser_indexes a ,user_ind_columns b
where a.table_name=b.table_name and a.index_name = b.index_name
and a.table_owner=upper('SAPSR3') and a.table_name='ANLU' order by a.uniqueness desc;
如果存在索引，并且选择性足够好，虽然不是主键列，但是可以直接使用keycols指定；
如果不存在索引？ 或者索引的选择性不够好，可以新增一个选择性好的索引，随后使用keycols进行指定。
！！！ 缺陷或者风险在于，因为是非主键，及时选择性好，如果存在null值,还是无法使用该索引，OGG 进程对应数据库session 执行SQL还是慢。 但是正常情况下可以解决该问题。
并且存在一个疑问，如果假设源端修改了一条记录，但是选择性好的column_name 在目标端对应2条记录，并且ogg复制进程使用keycols参数指定？ 那么目标端是修改更新2条记录？ 还是1条记录呢？？？
2.1 定位OGG复制进程在oracle数据库中的Session
$ps -ef|grep RP10 PID 20 RP10 ······ $ps -ef|grep 20 PID 60864 LOCAL=NO [OGG Session process]
SQL> select s.sid,s.serial#,sql_id,p.program from v$process p,v$session s where p.addr=s.paddr and p.spid=60864;
SID SERIAL# SQL_ID PROGRAM
------------------------------------------------
2276 37851 2j664 oracle@cc (TNS V1-V3)
2.2 定位造成OGG复制进程延迟过高的SQL
通过ash视图，查询1天内这个OGG进程 session ，都在执行什么sql ,event信息。 ash视图间隔1s采样 active session 1次。因此捕捉到的次数越多，说明消耗花费的时间越多。
select sql_id,event,BLOCKING_SESSION,CURRENT_OBJ#,count(*) from v$active_session_history where SAMPLE_TIME>sysdate-1 and SESSION_ID=2276 and SESSION_SERIAL#=37851
group by sql_id,event,BLOCKING_SESSION,CURRENT_OBJ# order by 5,4;
SQL_ID EVENT BLOCKING_SESSION CURRENT_OBJ# COUNT(*)
------------- ---------------------------------------------------------------- ---------------- ------------ ----------
1n7zz8wb86jpw 0 16
088mh1tws6wtm -1 19
2jt8ttg6b42b4 0 75
9y2087cvvr4r9 0 115
1n7zz8wb86jpw -1 4575
2cc4 -1 18877
9y2087ccc9 -1 62497
10 rows selected.
select * from table(dbms_xplan.display_cursor('2ccb4'));
606
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
