恢复Controlfile文件的疑问

ccsnmoracle

原帖由 Yong Huang 于 2010-2-26 02:55 发表
> DATA BUFFER往下写的9个条件
...
> 5.rac ping request

It's better to tell us what version these conditions apply to, and of course the source of this information unless it's from your own research. "RAC ping request" sounds like pre-9i i.e. OPS environment.

Just found it. This may be the source that many people are referencing:
http://www.oracle.com/technology ... tectureOverview.pdf

It has a copyright timestamp of 2001. I don't think Oracle9i was released in that year. But some concepts mentioned in the article are definitely newer than 8i, such as sga_max_size (beginning with 9iR2 I think). I think "RAC ping request" is a mistake. It's better to say "OPS (8i and before) ping request". OPS does not have cache fusion so ping request needs a write.

Additionally, any segment checkpoint, not just that before you drop or truncate table, but also that before Oracle scans a table or index in parallel, causes DBWn to write.

Yong Huang

还有segment checkpoint呀。

Yong Huang大师，能给我点资料不，关于checkpoint的。我想系统地学习一下。。。太乱了。。。。

我发现checkpoint的分类不少呀。。。。
有Normal Checkpoint,Tablespace Checkpoint,还有segment checkpoint...

从程度上划分，又分什么Complete Checkpoint和Incremental Checkpoint，Incremental Checkpoint是每三秒执行一次吗？


另外，我觉得。。。。（我的胡思乱想）
checkpoint事件的核心思想是将所谓的Dirty Blocks同步到数据文件中，以达到降低IO竞争的目的。而为了减少INSTANCE RECOVER的时间，Incremental Checkpoint和
Checkpoint Queue就被引入了。

我觉得我对这个Dirty Blocks的概念很重要，我个人的看法是，Dirty Blocks就只是特指由DML操作所引起的那些改变。

一个数据库，在她的生命周期中，像建立数据空间这样的操作能有几次呢？？比起DML操作的数量来，少的可怜吧。。。
所以我觉得，所有的DDL操作全都是立刻执行的，即立刻会修改数据文件，修改数据字典和控制文件等。

但也可能是这样的，比如我执行了CREATE TABLE,这势必会引起对数据字典的修改，数据字典也是保存在DATAFILE中的，所以要修改DATAFILE的内容，
这样一来就产生了Dirty Blocks。这个新的Dirty Blocks就进入了checkpint队列，等待被同步。如果这个假设成立的话，以上的那就错了。。。。


还是请大家帮帮忙，给讲讲吧。

jinqibingl

dirty blocks 应该是被修改过的数据块，被临时放置在lru中，等待被写入。而DDL操作尽管会引发字典的修改，不过这个字典修改的dirty blocks好像是在字典缓冲区里面，这个缓冲区和数据缓冲区的机制应该是略有不同吧。

Yong Huang

The concept of "segment checkpoint" is never officially acknowledged by Oracle. But you see it mentioned by a lot of people. For example, Jonathan Lewis uses this term in his book and many of his posted messages.

Note that not all dirty block writing is considered as checkpoint.

Yong Huang

gszoracle

如果有数据文件，log
restore,recover应该是可以的

jametong

原帖由 Yong Huang 于 2010-2-27 02:05 发表
The concept of "segment checkpoint" is never officially acknowledged by Oracle. But you see it mentioned by a lot of people. For example, Jonathan Lewis uses this term in his book and many of his posted messages.

Yong Huang

segment checkpoint 应该就是一般所说的Fast Object Checkpoint吧,,在并行读取或者truncate table/drop table 操作时会触发相应对象(segment)的checkpoint.

Note that not all dirty block writing is considered as checkpoint.

当一个事务修改的数据量非常大时,或者一个对象相关的blocks操作Buffer Cache的一定比例(我印象中是10%)时,或者没有足够的free buffer来放置新的block时都会触发DBWR将部分脏块将Flush到数据文件,,而这些脏块的写操作一般都与checkpoint没有什么关系..

[ 本帖最后由 jametong 于 2010-3-21 18:37 编辑 ]

Yong Huang

> segment checkpoint 应该就是一般所说的Fast Object Checkpoint吧

I think segment checkpoint even existed before 10gR2. But fast object checkpoint probably starts to appear in 10gR2 (controlled by _db_fast_obj_ckpt). Whether we can say they're exactly the same thing I'm not sure.

> 当一个事务修改的数据量非常大时,或者一个对象相关的blocks操作Buffer Cache的一定比例(我印象中是10%)时...

You remembered that number (10% of buffer cache) correctly. Those dirty blocks are placed on modified block lists. See
http://yong321.freeshell.org/oranotes/ModifiedBlockList.txt

Yong Huang

sqysl

记得9I的发布是在2001年。

hqsh

3q3q3q3q3q3q3q