Oracle是否需要保证同一个事务的改变向量顺序写出呢？

Alex螺丝钉

Alex螺丝钉

Alex螺丝钉

Alex螺丝钉

Alex螺丝钉

Alex螺丝钉

就这么多了。

Yong Huang

> 在private strand出现之前，redo record在pga生成，然后被server process从PGA复制到logbuffer中的shared strand ...
> 在private strand出现后，private strand在shared pool中创建

You're right. I should emphasize that I was only referring to the case with private strands. When Oracle does not use private strands, documentation clearly says
"The database processes copy redo entries from the user memory space to the redo log buffer in the SGA."
http://docs.oracle.com/database/121/CNCPT/memory.htm#CNCPT1225

> 我认为一个redo record应该使用的是连续的“buffer”，不可能出现楼主说的那种情况

I agree that a redo record is continuous (i.e. its change vectors occupy a contiguous area of memory in redo log buffer). But he (楼主) shows that a transaction, not the redo records of a transaction, has change vectors scattered in multiple log buffers. That's reasonable. If he truly meant redo records instead of transactions in the picture, he would sure be wrong.

Alex螺丝钉

Yong Huang 发表于 2015-10-22 23:48
> 在private strand出现之前，redo record在pga生成，然后被server process从PGA复制到logbuffer中的shared ...

谢谢版主的热心解答，想想，死扣这个东西也没啥意义。

Alex螺丝钉

anlinew 发表于 2015-10-21 09:32
这种内部实现机制，因为没法看到源代码，只能靠猜测
如果实在想搞明白，可以尝试dtrace试试看
还有个办法 ...

谢谢版主，我也不想搞明白这个东西了~~

Alex螺丝钉

zergduan 发表于 2015-10-21 13:34
我认为首先redo record (由多个cv组成）是先生成在pga中的，然后才被写入logbuffer中~
并且再写入的过程中 ...

thanks~~