RAC1与RAC2数据不一致

Yong Huang

When you're connecting to database account A in sqlplus and you type connect B, your session to A is disconnected and the transaction (if there was one) is committed.

Yong Huang

tendba

抱歉，发现一个自己的一个大漏洞。autocommit是针对每个命令的。

kenni

没commit就去查了？

tendba

原帖由 kenni 于 2010-12-5 18:12 发表
没commit就去查了？

commi了，所以lz才说只是偶尔才发生这样的现象。建立第二个连接的时候也是退出低一个连接的时候，自动提交最后一个transaction.

ts_keta

commit_wait参数设置成了nowait导致的。

Yong Huang

> commit_wait参数设置成了nowait导致的。

ts_keta, are you saying you found the root cause of your problem and it was because your commit_write (there's no commit_wait) was set to nowait? Is it explicitly set? According to
https://support.oracle.com/CSP/m ... =BUG&id=7142180

"In 11.1.0.6 and 10.2.0.4 the default commit behavior has changed so that
control is immediately being returned back to the client instead of waiting
for log buffer to be written to online log file. ( we are no longer waiting
for log file sync ) ... The default behavior is now that of commit_write=nowait, instead of wait."

So if you don't explicitly set it, it's nowait by default. Can you verify with many repeated tests as follows?

* Set commit_write to wait: you'll never see the problem you're reporting here.
* Don't set it or set it to nowait: some of your repeated tests show the problem.

Thank you.

Yong Huang

ts_keta

连接到:
Oracle Database 11g Enterprise Edition Release 11.1.0.6.0 - Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options


SQL> sho parameter commit_wait

NAME                                 TYPE        VALUE
------------------------------------ ----------- ------------------------------
commit_wait                          string      WAIT
SQL>


I'm sure that the promble was cased by this parameter.
session 1: control is immediately being returned back to the client instead of waiting
for commit, it means that the transaction of session 1 is not complete
session 2:so the select statement return no data

版主你有ORACLE的参数大全吗

Yong Huang

ts_keta,

Thanks for the correction. I didn't know till now 11g introduced commit_wait to replace commit_write.

All I can find is just Oracle Reference manual. MOS (support.oracle.com) doesn't have a good article describing all these changes. Looks like somebody can write a good article on this topic. Do you want to try?

Yong Huang

wolfop

commit_wait=nowait岂不是是类似于INFORMIX的bufferlog？
通过牺牲durable提升性能？

AndyChan

路过。
学习一下commit_wait的作用