sql解析中参数MAXOPENCURSORS, HOLD_CURSOR, and RELEASE_CURSOR 的解释

WHERE clause predicates are sent as string literals. If you use precompilers to develop the application, then make sure to reset the parameters MAXOPENCURSORS, 

HOLD_CURSOR, and RELEASE_CURSOR from the default values before precompiling the application

我们说的游标概念比较复杂,它可以是客户端程序中的游标,服务进程中的私有游标,以及服务器端共享池里的共享游标。假设一个游标被打开了,一般来说它的共享游标信息(包括执行计划,优化树等)总是会在SQL
AREA里,无需再次软/硬解析。

参数文件中的连个游标相关的值:

SQL> show parameter CURSORS

NAME     TYPE
VALUE
------------------------------------ ----------- ------------------------------
open_cursors     integer300
session_cached_cursors     integer50

open_cursors:指明每个session最多能够打开的cursors的数量。

SESSION_CACHED_CURSORS:是Oracle中的一个初始化参数(修改必须重启实例),指定了每个会话缓存的游标上限(保留在PGA中);客户端程序中open
cursor的要求仍会被传递给服务进程,服务进程首先扫描自身缓存的游标信息,如果命中则可以避免软解析,也有人称它为“软软解析”。

如果Open_cursors设置太小,对系统性能不会有明显改善,还可能触发ORA-O1000:m~imum open CUrsOrs exceeded.的错误。如果设置太大,则无端消耗系统内存。我们可以通过如下的sql语句查看你的设置是否合理:

SELECT MAX(A.VALUE) AS HIGHEST_OPEN_CUR, P.VALUE AS MAX_OPEN_CUR  
      FROM V$SESSTAT A, V$STATNAME B, V$PARAMETER P  
     WHERE A.STATISTIC# = B.STATISTIC#  
       AND B.NAME = 'opened cursors current'  
       AND P.NAME = 'open_cursors'  
     GROUP BY P.VALUE;  

HIGHEST_ OPEN CUR是实际打开的cursors 的最大值,MAX_OPEN_ CUR是参数Open_cursors的设定值,如果二者太接近,甚至触发eRA一01000错误,那么你就应该调大参数Open_cursors的设定值。

HOLD_CURSOR:是预编译程序中的一个参数,它指定了私有游标是否应该该被缓存,这里不做展开。

在分析工具tkprof中hard parse与soft parse被同等对待,都被认为是parse;软解析即会造成parse总数上升。

软解析避免了硬解析过程中的几个步骤,但仍包括了初始化的语法,语义解析并计算语句HASH值与SQL AREA中已有语句进行对比;若匹配则查询优化等昂贵的操作得以避免。

另请注意,10053事件仅在硬解析过程中被触发。

What are the Oracle Precompiler options HOLD_CURSOR and RELEASE_CURSOR and
how do they affect the performance of a PCC program?

First of all, these options apply to implicit and explicit cursors in all
precompiled languages except Pro*Ada. They apply ONLY to implicit cursors associated with INSERT, UPDATE, DELETE, or a single row SELECT in
Pro*Ada. An explicit cursor in a program is one explicitly created with EXEC SQL DECLARE C1 CURSOR FOR...

This bulletin discusses what these options do internally and how changing them
affects program performance.

HOLD_CURSOR is a precompiler command line parameter.

Summary Operation:
~~~~~~~~~~~~~~~~~~

NB: Both the soft and hard parse register as a parse in tkprof. If
the cursor is open and it is not in the SQL_AREA then it clearly has to parse it (not shown in the diagram!)

The HOLD_CURSOR and RELEASE_CURSOR Options
------------------------------------------
.
Following will be a discussion of what these options do internally and how
changing them affects program performance.

What exactly do we mean by CURSOR? Unfortunately, we mean two different
things:

1. The program cursor - A data structure associated with a SQL statement.

A program cursor is declared for each SQL statement that the  precompiler
finds in your program. For the statements

EXEC SQL DECLARE SEL_EMP_CURS CURSOR FOR...
EXEC SQL INSERT...

PCC will declare two program cursors, say c1 and c2.
2. The Oracle cursor (also called the context area) - The workarea
created dynamically at run time; 

this area contains the  parsed
statement, the addresses of the host variables, and  other information necessary to execute the SQL statement.

These two cursors are linked together via the cursor cache. The initial
size of the cursor cache is determined by the MAXOPENCURSORS option.

The following diagram illustrates the relationship described above after
an insert and an update have been executed in your program:

How are the HOLD_CURSOR and RELEASE_CURSOR options related to this view?

The HOLD_CURSOR option deals with the link between the program cursor and
its cache entry.

The RELEASE_CURSOR option deals with the link between the Oracle cursor
and the cache entry.

For SQL statements that are FREQUENTLY RE-EXECUTED, the bottom line is this:
if you want to maximize performance, make sure these SQL statements stay "glued" to their respective Oracle cursor.

What does it mean when a SQL statement is "glued" to its Oracle cursor? 

It means that both links between the SQL statement and its Oracle
cursor are made permanent.

Why would you want to keep a statement "glued" to its context area?
Because the context area contains the parsed statement and other information
necessary to execute the statement, such as the addresses of the host variables. Maintaining access to this information makes subsequent
execution of the statement much faster.

How do you "glue" a statement to a cache entry? 

By correct use of the  HOLD_CURSOR
and RELEASE_CURSOR options via the PCC command line or inline with EXEC ORACLE OPTION(...).

For instance, with HOLD_CURSOR=YES as the Oracle option, a cache entry cannot
be flagged for reuse. This has important implications. If all cache
entries have been used up and a new cache entry is needed for a new SQL statement such that the number of cache entries would now exceed
the number specified in MAXOPENCURSORS, Oracle will use the first cache entry marked reuseable.

For example, in the above diagram, if the cache entry C(1) is marked reusable,
and the program is about to execute the EXEC SQL SELECT...
(program cursor P(MAXOPENCURSORS+1), and the number of cache entries
in use already equals MAXOPENCURSORS, cache entry C(1) and its Oracle
cursor will now be linked to the select statement. A subsequent execution
of the insert statement would require pre-empting a cache entry and its Oracle cursor from another SQL statement and performing a reparse.

Correspondingly(相应的), with the default RELEASE_CURSOR=NO as the Oracle option,
the link between the cache entry and the Oracle cursor (the con-
text area) is maintained after the statement is executed so that the  parsed
statement and, more importantly, the allocated memory stay available.

The freeing up of this memory by RELEASE_CURSOR=YES means that the  next
statement that gets linked to this cache entry will require an  expensive reallocation of memory in addition to a reparse. Ugh! Why  would
anybody want RELEASE_CURSOR=YES? We will see later on.

Program cursor - - - - - [ Cursor cache entry ] - - - - - Oracle for
SQL statement cursor
                         ^                                            ^
HOLD_CURSOR=YES                     RELEASE_CURSOR=NO
program cursor is permanently           cache entry maintains the
linked to its cache entry.                   address of its context area.

So the HOLD_CURSOR option is intimately tied to the MAXOPENCURSORS option.
What exactly is the MAXOPENCURSORS option? First of all, MAX-
OPENCURSORS is a misnomer(用词不当). It should more appropriately be called INITIAL_CURSOR_CACHE_SIZE.
(Okay, so it's a mouthful.) Anyway, if
all cursor cache entries are currently marked "not reusable" either because
of the HOLD_CURSOR option or because the associated statement is currently being executed (an explicitly opened cursor is still being
fetched on and hasn't been closed yet), then a request for a new cursor will actually result in the extension of the cursor cache at runtime
(i.e. if MAXOPENCURSORS=10, and all 10 entries are active,then an 11th will be created). 

Just letting the precompiler reuse the oldest
cache entry won't always work, as the following example illustrates: Imagine the case where the user has ten explicitly declared  cursors
opened, and wants to execute an eleventh. If the program actually reuses the oldest program cursor, the user would lose his current
position in the first cursor and would not be able to fetch from it anymore.

By the way, if an eleventh cache entry is created, when that cursor is closed
the eleventh entry is not removed. Setting MAXOPENCURSORS low saves memory, but causes potentially expensive dynamic allocations of  new
cache entries if they're needed. Setting it high assures quick  execution, but may use more memory than necessary.

What if a statement is not executed repeatedly in a program? 

Then you could go with the other
options HOLD_CURSOR=NO and RELEASE_CURSOR=YES.   With the HOLD_CURSOR=NO option, the link between a program cursor and  its
cache entry is not permanent. The cache entry is automatically  marked reusable in case it is needed. With the RELEASE_CURSOR=YES option,
the Oracle cursor (the context area) is automatically freed and the parsed statement lost. A reason you might use this option is if  you
are limited by the number of Oracle cursors (MAXOPENCURSORS) at your site due to memory issues. You may want to incur the cost of reallocating
memory and reparsing in order to manage memory more effectively.

An advantage of setting RELEASE_CURSOR=YES is that until the link be-
tween the cache entry and the Oracle cursor (context area) is removed,
ORACLE keeps parse locks on any tables referenced in the SQL state-
ment. These parse locks prevent other users and you from ALTERing or
DROPping the tables (does ORA-0057 sound familiar?). Also, in Version
5, it will free up the read-consistent image of the referenced tables
stored in ORACLE's Before Image file.

What do we mean when we say that RELEASE_CURSOR=YES takes precedence
over HOLD_CURSOR=YES? With RELEASE_CURSOR=YES, the link between the
Oracle cursor and the cache entry is cut and the Oracle cursor is
freed (closed), so even if your program cursor is permanently linked
to the cache entry because HOLD_CURSOR=YES, you will still have to re-
allocate memory and reparse the statement. So subsequent executions
of a statement don't benefit from the HOLD_CURSOR=YES option because
RELEASE_CURSOR=YES.

For programmers experienced with OCI, here's the OCI equivalent of
what's happening: 

#define MAXOPENCURSORS 5 

char     *sql_stmts[10];
curs_def cursor[MAXOPENCURSORS]; 

oopen(cursor[0],...);
osql3(cursor[0],...,sql_stmts[0],...);  

An example of a "cache entry" being linked to another SQL statement
later on in the program is as follows: 

osql3(cursor[0],...,sql_stmts[5],...);  

I am forced to reuse one of my "cache entries" to execute the sixth
SQL statement. 

An example of a context area being freed is:  

oclose(cursor[0]);      

Reusing cursor[0] would require another oopen() and another osql3()--
another dynamic allocation of memory and another reparse. 

Conclusion
---------- 

As a programmer, you will get the most from these options by using
them selectively inline rather than specifying them as options at pre-
compile time.

Reference:

Pro*C/C++ Precompiler Programmer's Guide
时间: 2024-09-14 00:13:34

sql解析中参数MAXOPENCURSORS, HOLD_CURSOR, and RELEASE_CURSOR 的解释的相关文章

SQL 数据库中的存储过程的参数问题

  1.SQL 数据库中的存储过程的参数问题 怎么将SQL数据库中的存储过程中的参数既作为输出变量又作为输出变量? [sql] view plaincopy --drop proc proc_test --go create proc dbo.proc_test @in int, @out int out, @in_out int output as select @out = @in + @in_out, --1 + 2 = 3 @in_out = @out + 1 --3 + 1 = 4 g

c++-如何利用C++来解析SQL语句中的select语句?

问题描述 如何利用C++来解析SQL语句中的select语句? SELECT * FROM stu WHERE ((Name = "Jack") AND ((Age > 18) OR (Height > 170))) 这是一条SQL语句,怎么用C++解析它,并从内存数据中输出结果?求大神给点思路即可.... 解决方案 你是要解析?还是要执行? 如果要执行SQL语句,需要学习ADO编程,可以百度一下vc ado编程. 解决方案二: 有c++和数据库对接的各种库的,mysql的

.net中参数从aspx.cs中传接到aspx中作为SQL语句参数???

问题描述 .net中参数从aspx.cs中传接到aspx中作为SQL语句参数???在detail.aspx.cs中我定义了全局变量protectedintno;在detail.aspx.cs进行处理后打算传到detail.aspx界面中作为访问数据库的参数,以下是detail.aspx中的一段代码:<asp:SqlDataSourceID="SqlDataSource1"runat="server"ConnectionString="<%$Co

Sql Server中Substring函数的用法实例解析_MsSql

SQL 中的 substring 函数是用来抓出一个栏位资料中的其中一部分.这个函数的名称在不同的资料库中不完全一样: MySQL: SUBSTR( ), SUBSTRING( ) Oracle: SUBSTR( ) SQL Server: SUBSTRING( ) SQL 中的 substring 函数是用来截取一个栏位资料中的其中一部分. 例如,我们需要将字符串'abdcsef'中的'abd'给提取出来,则可用substring 来实现: select substring('abdcsef'

简单对比MySQL和Oracle中的一个sql解析细节

SQL的语法解析器是一个很强大的内置工具集,里面会涉及到很多的编译原理的相关知识,语法分析,词法分析..一大堆看起来很理论的东东,不过看起来枯燥之余,它们的价值也更加明显. 借用一下网络中的原话:如果我们考究一下历史,就会发现很多被称为程序设计大师的人都是编译领域的高手.写出第一个微型机上运行的Basic语言的比尔盖茨,设计出Delphi的Borland的"世界上最厉害的程序员", Sun的JAVA之父, 贝尔实验室的C++之父 起点提得有些高了,今天和大家分享的案例是一个很简单的sq

Sql Server中Substring函数的用法实例解析

SQL 中的 substring 函数是用来抓出一个栏位资料中的其中一部分.这个函数的名称在不同的资料库中不完全一样: MySQL: SUBSTR( ), SUBSTRING( ) Oracle: SUBSTR( ) SQL Server: SUBSTRING( ) SQL 中的 substring 函数是用来截取一个栏位资料中的其中一部分. 例如,我们需要将字符串'abdcsef'中的'abd'给提取出来,则可用substring 来实现: select substring('abdcsef'

解析SQL语句中Replace INTO与INSERT INTO的不同之处

本篇文章是对SQL语句中Replace INTO与INSERT INTO的不同之处进行了详细的分析介绍,需要的朋友参考下   REPLACE的运行与INSERT很相似.只有一点例外,假如表中的一个旧记录与一个用于PRIMARY KEY或一个UNIQUE索引的新记录具有相同的值,则在新记录被插入之前,旧记录被删除. 注意,除非表有一个PRIMARY KEY或UNIQUE索引,否则,使用一个REPLACE语句没有意义.该语句会与INSERT相同,因为没有索引被用于确定是否新行复制了其它的行. 所有列

解析sql语句中left

以下是对在sql语句中left_join.inner_join中的on与where的区别进行了详细的分析介绍,需要的朋友可以参考下   table a(id, type): id     type ---------------------------------- 1      1         2      1          3      2          table b(id, class): id    class -------------------------------

请教大虾:向SQL语句中传递参数的问题

问题描述 1.SqlConnectionconn1=newSqlConnection("database=charm;server=sql8;user=sa;pwd=8");2.this.mA=newSqlDataAdapter("selectgdcode,gdno,gddesc,gdunit,gdpricfromslgoodsql1",conn1);3.stringsql1="wheregdcode='"+combo1+"'"