oracle consistent gets,oracle构建一致性读

对于实际的业务系统，通常有一些热点的表，insert和delete的量非常大，这个时候就会发现一些查询语句的逻辑读比较偏高，

这时可能就是oracle在构建一致性块的进行的consistent

read。

下面做一个测试看下：

第一步准备数据：

create tabletest(

col1varchar2(12)

col2numberextvarchar2(4000)

);create index test_ind on test(user_id, col2);create sequence seq_test cache 200;

这样这样的表，我们假设有频繁的插入和删除操作，那么下面来测试一下select的逻辑读的情况。

开启两个session:

1，创建表保存snapshot

在session1：create table prefix_stats tablespace IW_ACCOUNT_LOG_01 as select * from v$sesstat where sid=&1;

2，在session2查询

select * from (select * from test t where col1 = ‘xpchild001‘ order by trans_log_id) where rownum <= 200;

3，在session1监控session2的统计信息

select *

from (selectt.name,

pre.valueaspre,

suf.valueassuf,

(suf.value- pre.value) asdifffromprefix_stats pre, v$sesstat suf, v$statname twhere pre.sid =suf.sidand pre.STATISTIC# =suf.STATISTIC#and pre.STATISTIC# =t.STATISTIC#) tmpwhere tmp.diff > 0

order by tmp.diff descName PRE SUF DIFF---------------------------------------------------------------------- ---------- ---------- ----------

session pga memory max 957208 1153816 196608session pga memory957208 1153816 196608bytes sent via SQL*Net to client 6692 37013 30321redo size0 8256 8256session logical reads52 1508 1456consistent getsfrom cache 52 1508 1456consistent gets52 1508 1456bytes received via SQL*Net from client 4385 5639 1254consistent gets- examination 21 1253 1232data blocks consistent reads- undo records applied 0 920 920consistent changes0 920 920bufferis not pinned count 17 222 205

table fetch by rowid 6 206 200bufferis pinned count 0 197 197CR blocks created0 160 160callsto kcmgas 0 160 160db block changes0 120 120redo entries0 120 120cleanout- number of ktugct calls 0 120 120cleanoutsand rollbacks - consistent read gets 0 120 120immediate (CR) block cleanout applications0 120 120nowork - consistent read gets 19 83 64heap block compress0 51 51rollbacksonly - consistent read gets 0 40 40shared hash latch upgrades- no wait 0 5 5

user calls 28 33 5

execute count 21 23 2DB time0 2 2parsecount (total) 22 24 2sessioncursor cache count 16 17 1CPU usedwhen call started 0 1 1recursive calls92 93 1parsecount (hard) 0 1 1sessioncursor cache hits 4 5 1CPU usedby this session 0 1 1

这一次的查询，返回记录200条。table fetch by rowid=200;

1，逻辑读session logical reads=consistent gets(一致读)+db block

gets(当前读);

这个sql只有一致性读session logical reads=consistent gets=1456

2，构建一致性读应用回滚记录统计：data blocks consistent reads(undo records

applied)：920

等价于consistent changes。

3，需要回滚或者块清除产生的一致性读(这里边没有回滚，只可能有块清除)cleanouts and rollbacks - consistent read

gets：120

跟db block changes=120一致，这里进行了块清楚，从而改变了db block。

4，构建一致性读clone的块数：CR blocks created=160

5，块清除产生的redo：redo size 8256