consistent gets如何理解

吴_飚

我简单的理解的，不是physical reads 就是 consistent gets。
physical reads 是啥大家都懂的。。

sqysl

吴_飚 发表于 2013-1-3 15:12
我简单的理解的，不是physical reads 就是 consistent gets。
physical reads 是啥大家都懂的。。

这个理解似乎有点问题。

一致性读的计数。
session  发出的对某个对块的一致性读的请求？

Yong Huang

> 不是physical reads 就是 consistent gets。

No. The opposite of physical reads is memory reads. Their sum is equal to the sum of consistent reads and current block reads. But the two components in each of the two sums cannot be paired.

sliangzi

这个SQL执行时的逻辑读。

lpcet4

就我的理解是 如果一个表刚刚创建好，没有执行过任何的DML语句，如查这里我执行一个SELECT 的查询语句，我觉得consistent gets应该0才对，因为现在只是在执行一个简单的查询，并且在同一时间并没有对这个表出现过DML语句的修改修，所以不应该出现构造CR块，但是实际中，往往consistent gets并不为0。我的意思是在某一个阶段之前并没有对某一个表做过DML操作，如果执行查询一个表的话，就不会应该出现CR现像，可能我把consistent gets和CR块混了。

Yong Huang

> 这个SQL执行时的逻辑读。

No. Logical reads include consistent reads *and current mode reads*.

> 就我的理解是 如果一个表刚刚创建好，没有执行过任何的DML语句，如查这里我执行一个SELECT 的查询语句，我觉得consistent gets应该0才对

No. The word "consistent" does not mean the read must involve CR block read or creation. It's just the name given to the mode of read.

stevehai812

Yong Huang 发表于 2012-12-10 22:43
Bad grammar: 指那是那些读内存操作

Consistent gets means gets (reads) in consistent mode. The read ...

you just said "he read is read of a buffer in buffer cache (memory)". when I execute ALTER SYSTEM FLUSH BUFFER_CACHE; The FLUSH BUFFER_CACHE clause lets you clear all data from the buffer cache in the system global area (SGA).Then I try to execute the SQL again,why does the 'Consistent gets' still have a non-zero value?  Thx.

stevehai812

stevehai812 发表于 2014-3-25 11:08
you just said "he read is read of a buffer in buffer cache (memory)". when I execute ALTER SYSTEM  ...

sorry, my mistake. The data should be read to the memory whenever I FLUSHED THE BUFFER CACHE or not!

Yong Huang

To stevehai812,

I see why you had the question ("you just said '[t]he read is read of a buffer in buffer cache (memory)'..."). I checked my message in msg#2. That sentence should not be read alone, otherwise it would indeed be mis-interpreted. What I meant is that all regular (i.e. excluding direct path read) disk reads of a table or index are reads from disk to buffer cache. Consistent read counts the buffer blocks needed to satisfy the SQL. So in this sense you can ignore the first phase of the whole process (disk read) and just count the second phase (buffer cache read).