偷窥Data block 的物理结构

venee_lee

frre=-1
fsbo=0x14                ==> free space begin offset
fseo=0x1f9b              ==> free space end offset
avsp=0x1f83              ==> available space in the block
tosp=0x1f83              ==> total available space when all transactions commit

怎么去理解这几个参数。我对这里的offset总觉得不清楚。是以什么地方做为计算的起始点？

fsbo = hsiz？
frre 似乎都是-1。

Yong Huang

原帖由 venee_lee 于 2009-4-7 01:24 发表
fsbo=0x14                ==> free space begin offset
fseo=0x1f9b              ==> free space end offset
...
怎么去理解这几个参数。我对这里的offset总觉得不清楚。是以什么地方做为计算的起始点？

If I remember right, the starting position is still the beginning of the block. So the block looks like this (assuming 8192 i.e. x2000 block size):

1 start 0 ----------
2 fsbo x14 ---------
3 fseo x1f9b ---------
4 end x2000 ---------

Data is in [3-4]. If there's new data coming in now, it's at 3 going toward 2. Block header is in [1-2].

I could remember wrong. Need to refresh my memory.

Yong Huang

venee_lee

原帖由 Yong Huang 于 2009-4-8 03:17 发表


If I remember right, the starting position is still the beginning of the block. So the block looks like this (assuming 8192 i.e. x2000 block size):

1 start 0 ----------
2 fsbo x14 ---------
3 fseo x1f9b ---------
4 end x2000 ---------

Data is in [3-4]. If there's new data coming in now, it's at 3 going toward 2. Block header is in [1-2].

I could remember wrong. Need to refresh my memory.

Yong Huang

和我的理解差不多！
1   fsbo=0x14
2   fseo=0x1f9b
3   avsp=0x1f83
4   tosp=0x1f83
-------------------------------------------------------------
1                                         2@@@@@@@@@@
@ 区域是已用空间
1:可用空间的起始位置
2:可用空间的终点位置
  2之后的是已用位置
1-2之间就是可用空间
fseo-fsbo=0x1f9b-0x14=0x1f87
可块头上记录的是avsp=0x1f83
这4个字节被哪里占用了，还是我计算方法有问题。

可否断定：数据的存储是从块的底部开始的？

[ 本帖最后由 venee_lee 于 2009-4-8 08:59 编辑 ]

Yong Huang

Indeed, data comes in from the bottom of the block. I don't know how to explain the missing 4 bytes. Is it related to delayed block cleanout? Can you make sure to use a query (SELECT statement) to read this block and dump again? What was the recent operation on this block, delete, insert, or update?

I'll keep this in mind. But I don't have an answer for now.

Yong Huang

[ 本帖最后由 Yong Huang 于 2009-4-8 13:47 编辑 ]

venee_lee

原帖由 Yong Huang 于 2009-4-8 09:33 发表
Indeed, data comes in from the bottom of the block. I don't know how to explain the missing 4 bytes. Is it related to delayed blocked cleanout? Can you make sure to use a query (SELECT statement) to read this block and dump again? What was the recent operation on this block, delete, insert, or update?

I'll keep this in mind. But I don't have an answer for now.

Yong Huang

上面贴出来的是楼主的数据。

我在自己的库上做过两次测试，两个不同的表。
过程
insert 一次
commit

两次的结果还不一样，其中的一次正好，另外一次也少4个byte。弄不懂了！

d.c.b.a

可以看看row header那儿有没有浪费的, 删除4条记录,然后再删除中间的两条试试.

weier_breath

'所以，一个block中最多包括 db_block_size/11 条rows'

是不是应该这样算:
(db_block_size-90)/11 条rows

wangkxxe

看看精华帖学习中！

itpub_2000

mark

readeveryday

mark