资源池“内部”中的系统内存不足

时间:2009-10-08 13:30:17

标签: sql-server-2008 memory-leaks oledb linked-server

SQL Server 2008链接服务器和临时INSERT导致快速内存泄漏,最终导致服务器无响应,并以以下错误结束:

Msg 701, Level 17, State 123, Server BRECK-PC\SQLEXPRESS, Line 2
There is insufficient system memory in resource pool 'internal' to run this 
query.

Location:        qxcntxt.cpp:1052
Expression:      cref == 0
SPID:            51
Process ID:      1880

在重新启动SQL Server之前,服务器仍然没有响应。

正在使用的软件:

  • Windows Vista Ultimate 64位版本6001 SP1

  • Microsoft SQL Server 2008(SP1) - 10.0.2734.0(X64)2009年9月11日14:30:58版权所有(c)1988-2008 Microsoft Corporation在Windows NT上具有高级服务(64位)的Express Edition 6.0(Build 6001:Service Pack 1)

  • SQL Anywhere 11.0.1.2276中的SAOLEDB.11驱动程序

将最大服务器内存(MB)设置为2048无济于事。

将各种-g值(例如-g256;)添加到服务器启动参数没有帮助。

使用DBCC FREESYSTEMCACHE('ALL'),DBCC FREESESSIONCACHE和DBCC FREEPROCCACHE没有帮助。

将Cumnulative更新包4安装到SQL Server 2008 Service Pack 1没有帮助,即使它包含对涉及链接服务器使用的内存泄漏症状的修复。

从INSERT中分离SELECT ... ROW_NUMBER()OVER ...查询没有帮助。实验表明复杂的SELECT没有导致内存泄漏,INSERT就是这样。

更改代码以使用ad-hoc“INSERT INTO OPENROWSET”语法而不是链接服务器没有帮助;下面的代码显示了链接服务器的使用情况。

sysinternals.com Process Explore实用程序显示内存使用情况与sqlserver.exe相关联,而不是SQL Anywhere OLEDB驱动程序SAOLEDB.11使用的DLL。

请注意,链接服务器(代理表)的SQL Anywhere版本可以正常工作,以便在单个事务中将190万行从SQL Server 2008表“拉”到SQL Anywhere 11数据库。这里显示的逻辑是尝试使用链接服务器功能“推”行;相同的方向,不同的语法。

代码如下;执行三次或四次EXECUTE copy_mss_t2后,RAM的4G耗尽:

EXEC sys.sp_configure 
   N'show advanced options',  
   N'1'
GO

RECONFIGURE WITH OVERRIDE
GO

EXEC sys.sp_configure
   N'max server memory (MB)',
   N'2048'
GO

RECONFIGURE WITH OVERRIDE
GO

EXEC sys.sp_configure
   N'show advanced options',
   N'0'
GO

RECONFIGURE WITH OVERRIDE
GO
EXEC master.dbo.sp_MSset_oledb_prop
   N'SAOLEDB.11',
   N'AllowInProcess',
   1
GO

sp_addlinkedserver
   @server = 'mem',
   @srvproduct = 'SQL Anywhere OLE DB Provider',
   @provider = 'SAOLEDB.11',
   @datasrc = 'mem_PAVILION2'
GO

EXEC master.dbo.sp_serveroption
   @server=N'mem',
   @optname=N'rpc',
   @optvalue=N'true'
GO

EXEC master.dbo.sp_serveroption
   @server=N'mem',
   @optname=N'rpc out',
   @optvalue=N'true'
GO

sp_addlinkedsrvlogin
   @rmtsrvname = 'mem',
   @useself = 'false',
   @locallogin = NULL,
   @rmtuser = 'dba',
   @rmtpassword = 'sql'
GO

CREATE PROCEDURE copy_mss_t2
   @from_row            BIGINT,
   @to_row              BIGINT,
   @rows_copied_count   BIGINT OUTPUT
AS

   SELECT *
     INTO #t
     FROM ( SELECT *,
                   ROW_NUMBER()
                      OVER ( ORDER BY sample_set_number,
                                      connection_number )
                   AS t2_row_number
             FROM mss_t2 ) AS ordered_mss_t2
    WHERE ordered_mss_t2.t2_row_number BETWEEN @from_row AND @to_row;

   SELECT @rows_copied_count = COUNT(*)
     FROM #t;

INSERT INTO mem..dba.sa_t2
SELECT sampling_id,
       sample_set_number,
       connection_number,
       blocker_owner_table_name,
       blocker_lock_type,
       blocker_owner_name,
       blocker_table_name,
       blocker_reason,
       blocker_row_identifier,
       current_engine_version,
       page_size,
       ApproximateCPUTime,
       BlockedOn,
       BytesReceived,
       BytesSent,
       CacheHits,
       CacheRead,
   "Commit",
   DiskRead,
   DiskWrite,
   FullCompare,
   IndAdd,
   IndLookup,
   Isolation_level,
   LastReqTime,
   LastStatement,
   LockCount,
   LockName,
   LockTableOID,
   LoginTime,
   LogWrite,
   Name,
   NodeAddress,
   Prepares,
   PrepStmt,
   QueryLowMemoryStrategy,
   QueryOptimized,
   QueryReused,
   ReqCountActive,
   ReqCountBlockContention,
   ReqCountBlockIO,
   ReqCountBlockLock,
   ReqCountUnscheduled,
   ReqStatus,
   ReqTimeActive,
   ReqTimeBlockContention,
   ReqTimeBlockIO,
   ReqTimeBlockLock,
   ReqTimeUnscheduled,
   ReqType,
   RequestsReceived,
   Rlbk,
   RollbackLogPages,
   TempFilePages,
   TransactionStartTime,
   UncommitOp,
   Userid,
   previous_ApproximateCPUTime,
   interval_ApproximateCPUTime,
   previous_Commit,
   interval_Commit,
   previous_Rlbk,
   interval_Rlbk
  FROM #t;

GO

DECLARE @rows_copied_count BIGINT
EXECUTE copy_mss_t2 1110001, 1120000, @rows_copied_count OUTPUT
SELECT @rows_copied_count
GO

EXECUTE create_linked_server
GO

DECLARE @rows_copied_count BIGINT
EXECUTE copy_mss_t2 1120001, 1130000, @rows_copied_count OUTPUT
SELECT @rows_copied_count
GO

EXECUTE create_linked_server
GO

这是SQL Server源表,包含190万行中大约1G的数据:

CREATE TABLE mss_t2 (
   sampling_id                       BIGINT NOT NULL,
   sample_set_number                 BIGINT NOT NULL,
   connection_number                 BIGINT NOT NULL,
   blocker_owner_table_name          VARCHAR ( 257 ) NULL,
   blocker_lock_type                 VARCHAR ( 32 ) NULL,
   blocker_owner_name                VARCHAR ( 128 ) NULL,
   blocker_table_name                VARCHAR ( 128 ) NULL,
   blocker_reason                    TEXT NULL,
   blocker_row_identifier            VARCHAR ( 32 ) NULL,
   current_engine_version            TEXT NOT NULL,
   page_size                         INTEGER NOT NULL,
   ApproximateCPUTime                DECIMAL ( 30, 6 ) NULL,
   BlockedOn                         BIGINT NULL,
   BytesReceived                     BIGINT NULL,
   BytesSent                         BIGINT NULL,
   CacheHits                         BIGINT NULL,
   CacheRead                         BIGINT NULL,
   "Commit"                          BIGINT NULL,
   DiskRead                          BIGINT NULL,
   DiskWrite                         BIGINT NULL,
   FullCompare                       BIGINT NULL,
   IndAdd                            BIGINT NULL,
   IndLookup                         BIGINT NULL,
   Isolation_level                   BIGINT NULL,
   LastReqTime                       TEXT NOT NULL DEFAULT '1900-01-01',
   LastStatement                     TEXT NULL,
   LockCount                         BIGINT NULL,
   LockName                          BIGINT NULL,
   LockTableOID                      BIGINT NULL,
   LoginTime                         TEXT NOT NULL DEFAULT '1900-01-01',
   LogWrite                          BIGINT NULL,
   Name                              VARCHAR ( 128 ) NULL,
   NodeAddress                       TEXT NULL,
   Prepares                          BIGINT NULL,
   PrepStmt                          BIGINT NULL,
   QueryLowMemoryStrategy            BIGINT NULL,
   QueryOptimized                    BIGINT NULL,
   QueryReused                       BIGINT NULL,
   ReqCountActive                    BIGINT NULL,
   ReqCountBlockContention           BIGINT NULL,
   ReqCountBlockIO                   BIGINT NULL,
   ReqCountBlockLock                 BIGINT NULL,
   ReqCountUnscheduled               BIGINT NULL,
   ReqStatus                         TEXT NULL,
   ReqTimeActive                     DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockContention            DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockIO                    DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockLock                  DECIMAL ( 30, 6 ) NULL,
   ReqTimeUnscheduled                DECIMAL ( 30, 6 ) NULL,
   ReqType                           TEXT NULL,
   RequestsReceived                  BIGINT NULL,
   Rlbk                              BIGINT NULL,
   RollbackLogPages                  BIGINT NULL,
   TempFilePages                     BIGINT NULL,
   TransactionStartTime              TEXT NOT NULL DEFAULT '1900-01-01',
   UncommitOp                        BIGINT NULL,
   Userid                            VARCHAR ( 128 ) NULL,
   previous_ApproximateCPUTime       DECIMAL ( 30, 6 ) NOT NULL DEFAULT 0.0,
   interval_ApproximateCPUTime       AS ( COALESCE ( "ApproximateCPUTime", 0 ) - previous_ApproximateCPUTime ),
   previous_Commit                   BIGINT NOT NULL DEFAULT 0,
   interval_Commit                   AS  ( COALESCE ( "Commit", 0 ) - previous_Commit ),
   previous_Rlbk                     BIGINT NOT NULL DEFAULT 0,
   interval_Rlbk                     AS  ( COALESCE ( Rlbk, 0 ) - previous_Rlbk ) )

以下是SQL Anywhere 11中的目标表:

CREATE TABLE sa_t2 (
   sampling_id                       BIGINT NOT NULL,
   sample_set_number                 BIGINT NOT NULL,
   connection_number                 BIGINT NOT NULL,
   blocker_owner_table_name          VARCHAR ( 257 ) NULL,
   blocker_lock_type                 VARCHAR ( 32 ) NULL,
   blocker_owner_name                VARCHAR ( 128 ) NULL,
   blocker_table_name                VARCHAR ( 128 ) NULL,
   blocker_reason                    TEXT NULL,
   blocker_row_identifier            VARCHAR ( 32 ) NULL,
   current_engine_version            TEXT NOT NULL,
   page_size                         INTEGER NOT NULL,
   ApproximateCPUTime                DECIMAL ( 30, 6 ) NULL,
   BlockedOn                         BIGINT NULL,
   BytesReceived                     BIGINT NULL,
   BytesSent                         BIGINT NULL,
   CacheHits                         BIGINT NULL,
   CacheRead                         BIGINT NULL,
   "Commit"                          BIGINT NULL,
   DiskRead                          BIGINT NULL,
   DiskWrite                         BIGINT NULL,
   FullCompare                       BIGINT NULL,
   IndAdd                            BIGINT NULL,
   IndLookup                         BIGINT NULL,
   Isolation_level                   BIGINT NULL,
   LastReqTime                       TEXT NOT NULL DEFAULT '1900-01-01',
   LastStatement                     TEXT NULL,
   LockCount                         BIGINT NULL,
   LockName                          BIGINT NULL,
   LockTableOID                      BIGINT NULL,
   LoginTime                         TEXT NOT NULL DEFAULT '1900-01-01',
   LogWrite                          BIGINT NULL,
   Name                              VARCHAR ( 128 ) NULL,
   NodeAddress                       TEXT NULL,
   Prepares                          BIGINT NULL,
   PrepStmt                          BIGINT NULL,
   QueryLowMemoryStrategy            BIGINT NULL,
   QueryOptimized                    BIGINT NULL,
   QueryReused                       BIGINT NULL,
   ReqCountActive                    BIGINT NULL,
   ReqCountBlockContention           BIGINT NULL,
   ReqCountBlockIO                   BIGINT NULL,
   ReqCountBlockLock                 BIGINT NULL,
   ReqCountUnscheduled               BIGINT NULL,
   ReqStatus                         TEXT NULL,
   ReqTimeActive                     DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockContention            DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockIO                    DECIMAL ( 30, 6 ) NULL,
   ReqTimeBlockLock                  DECIMAL ( 30, 6 ) NULL,
   ReqTimeUnscheduled                DECIMAL ( 30, 6 ) NULL,
   ReqType                           TEXT NULL,
   RequestsReceived                  BIGINT NULL,
   Rlbk                              BIGINT NULL,
   RollbackLogPages                  BIGINT NULL,
   TempFilePages                     BIGINT NULL,
   TransactionStartTime              TEXT NOT NULL DEFAULT '1900-01-01',
   UncommitOp                        BIGINT NULL,
   Userid                            VARCHAR ( 128 ) NULL,
   previous_ApproximateCPUTime       DECIMAL ( 30, 6 ) NOT NULL DEFAULT 0.0,
   interval_ApproximateCPUTime       DECIMAL ( 30, 6 ) NOT NULL COMPUTE ( COALESCE ( "ApproximateCPUTime", 0 ) - previous_ApproximateCPUTime ),
   previous_Commit                   BIGINT NOT NULL DEFAULT 0,
   interval_Commit                   BIGINT NOT NULL COMPUTE ( COALESCE ( "Commit", 0 ) - previous_Commit ),
   previous_Rlbk                     BIGINT NOT NULL DEFAULT 0,
   interval_Rlbk                     BIGINT NOT NULL COMPUTE ( COALESCE ( Rlbk, 0 ) - previous_Rlbk ),
   PRIMARY KEY ( sample_set_number, connection_number ) );

5 个答案:

答案 0 :(得分:8)

每次迭代后,您是否需要清空临时表#t?即在程序结束时添加TRUNCATE TABLE #t?我认为临时表#t在您的会话结束之前一直存在,直到存储过程结束。 SELECT INTO只是附加到现有的#t,但不会替换它。

另一件事是使用永久表而不是存储在tempdb #tables中的东西。

答案 1 :(得分:2)

问题是通过SQL Anywhere 11.0.1提供程序SAOLEDB.11使用链接服务器将数据插入声明为大于VARCHAR(8000)的目标列。这是一个简化的可重复的:

-- Overview: SQL Server 2008 suffers from a fatal memory leak
--    if an attempt is made to use a linked server and the 
--    SAOLEDB.11 provicer to copy data from SQL Server
--    to a SQL Anywhere 11.0.1 table that contains a single column
--    larger than VARCHAR ( 8000 ); i.e, a VARCHAR ( 8000 ) target
--    column is OK but VARCHAR ( 8001 ) is not. The actual string 
--    length is not an issue, nor is the fact that the SQL Server
--    column is declared as TEXT. The memory leak is faster if
--    there is more than one target column larger than VARCHAR ( 8000 ).
--    The server computer freezes and must be rebooted.
-- Msg 701, Level 17, State 123, Server BRECK-PC\SQLEXPRESS, Line 2
-- There is insufficient system memory in resource pool 'internal' to run this
-- query.
-- Location:        qxcntxt.cpp:1052
-- Expression:      cref == 0
-- SPID:            52
-- Process ID:      2044

---------------------------------------------------------
-- SQL ANYWHERE 11 on target computer
---------------------------------------------------------

-- Target:
-- HP Pavilion laptop, 4GHz Pentium 4, 2G RAM
-- Windows XP SP2
-- SQL Anywhere 11.0.1.2276

---------------------------------------------------------
-- Windows commands used to create and start in-memory database

/*
"%SQLANY11%\bin32\dbinit.exe"^
  mem.db

"%SQLANY11%\bin32\dbspawn.exe" -f^
  "%SQLANY11%\bin32\dbsrv11.exe"^
  -im nw^
  -o dbsrv11_log.txt^
  mem.db 

"%SQLANY11%\bin32\dbisql.com"^
  -c "ENG=mem;DBN=mem;UID=dba;PWD=sql;CON=mem-1"
*/

---------------------------------------------------------
-- Create target table with one single "long" column.

BEGIN
   DROP TABLE sa_target;
   EXCEPTION WHEN OTHERS THEN
END;

CREATE TABLE sa_target (
   primary_key      INTEGER NOT NULL PRIMARY KEY,
   string_column_1  VARCHAR ( 8001 ) NOT NULL );

---------------------------------------------------------
--- SQL SERVER 2008 on source (server) computer
---------------------------------------------------------

-- Server:
-- Desktop with Intel Core 2 Quad Q9450 2.66Ghz 4G RAM
-- Windows Vista Ultimate 64 bit build 6001 SP1
-- SQL Server 2008 Express 64 Service Pack 1 with cumulative update package 4:
--    Microsoft SQL Server 2008 (SP1) - 10.0.2734.0 (X64)   Sep 11 2009 14:30:58   
--    Copyright (c) 1988-2008 Microsoft Corporation  Express Edition with    
--    Advanced Services (64-bit) on Windows NT 6.0 <X64>    
--    (Build 6001: Service Pack 1) 
-- SAOLEDB.11 driver from SQL Anywhere 11.0.1.2276

---------------------------------------------------------
-- Windows command used to start osql.exe

/*
"c:\Program Files\Microsoft SQL Server\100\Tools\Binn\osql.exe"^
  -d main^
  -I^
  -l 10^
  -P j68Fje9#fyu489^
  -S BRECK-PC\SQLEXPRESS^
  -U sa
*/

---------------------------------------------------------
-- Create database.

USE master
GO

BEGIN TRY
   DROP DATABASE main;
END TRY
BEGIN CATCH
END CATCH;
GO

CREATE DATABASE main
ON PRIMARY
( NAME = main_dat,
   FILENAME = 'E:\data\main\main.mdf',
   SIZE = 2GB,
   FILEGROWTH = 200MB )
LOG ON
( NAME = 'main_log',
   FILENAME = 'E:\data\main\main.ldf',
   SIZE = 2GB,
   FILEGROWTH = 200MB )
GO

----------------------------------------------------------------------------
-- Configure SAOLEDB.11 provider.

USE master
go 

-- SAOLEDB.11 provider dlls registered via these Windows commands:
--    regsvr32 dboledb11.dll
--    regsvr32 dboledba11.dll

EXEC master.dbo.sp_MSset_oledb_prop N'SAOLEDB.11', N'AllowInProcess', 1
GO

-- If the following statement produces this message, it probably means 
-- that 'DisallowAdHocAccess' is already set to zero for SAOLEDB.11:
--
-- RegDeleteValue() returned error 2, 'The system cannot find the file specified.'

EXEC master.dbo.sp_MSset_oledb_prop N'SAOLEDB.11', N'DisallowAdHocAccess', 0
GO

----------------------------------------------------------------------------
-- THIS SECTION WAS NOT RUN.
-- Set up for OPENROWSET usage.
-- NOT REQUIRED for required for Linked Server usage.

USE master
GO

sp_configure 'show advanced options', 1
GO

RECONFIGURE
GO

sp_configure 'Ad Hoc Distributed Queries', 1
GO

RECONFIGURE
GO

----------------------------------------------------------------------------
-- Set up Linked Server usage.

USE main
GO

BEGIN TRY
   EXEC sp_droplinkedsrvlogin 
      @rmtsrvname = 'mem',
      @locallogin = NULL  
END TRY
BEGIN CATCH
END CATCH 
GO

BEGIN TRY
   EXEC sp_dropserver
      @server = 'mem'
END TRY
BEGIN CATCH
END CATCH 
GO

EXEC sp_addlinkedserver
   @server = 'mem',
   @srvproduct = 'SQL Anywhere OLE DB Provider',
   @provider = 'SAOLEDB.11',
   @datasrc = 'mem_PAVILION2' 
GO

EXEC master.dbo.sp_serveroption 
   @server=N'mem', 
   @optname=N'rpc', 
   @optvalue=N'true' 
GO

EXEC master.dbo.sp_serveroption 
   @server=N'mem', 
   @optname=N'rpc out', 
   @optvalue=N'true' 
GO

EXEC sp_addlinkedsrvlogin 
   @rmtsrvname = 'mem', 
   @useself = 'false', 
   @locallogin = NULL, 
   @rmtuser = 'dba', 
   @rmtpassword = 'sql' 
GO

----------------------------------------------------------------------------
-- Create and fill source table with 1 million rows.

USE main
GO

BEGIN TRY
   DROP TABLE mss_source; 
END TRY
BEGIN CATCH
END CATCH 
GO

CREATE TABLE mss_source ( 
   primary_key      INTEGER NOT NULL PRIMARY KEY,
   string_column_1  TEXT NOT NULL )
GO

BEGIN
   DECLARE @primary_key INTEGER 
   SELECT @primary_key = 1
   BEGIN TRANSACTION
   WHILE @primary_key <= 1000000 
   BEGIN
      INSERT mss_source VALUES (
         @primary_key,
         REPLICATE ( 'Some test data. ', 2 ) )
      SELECT @primary_key = @primary_key + 1
   END
   COMMIT
END
GO

SELECT COUNT(*) FROM mss_source
GO

-- 1000000

---------------------------------------------------------
-- Copy data to target table.

SELECT CURRENT_TIMESTAMP
GO

INSERT INTO mem..dba.sa_target 
SELECT * 
  FROM mss_source
GO

SELECT CURRENT_TIMESTAMP
GO

---------------------------------------------------------
-- Test 1 - Code as shown above.
-- FAILED
-- 
-- Started at 2009-10-12 10:06:33.393
-- 
-- A slow server memory leak began immediately.
-- The initial target insert rate was about 2000 rows per second.
-- Server RAM usage reached 3.82 GB, Physical Memory: 95%, Page File 16236M / 16288M
-- The server display became frozen.
-- The server became unresponsive to mouse input.
-- The target insert rate dropped below 1000 rows per second, but inserts continued.
-- The copy process reached 937,817 rows inserted on the target.
-- This dialog box appeared on the server: "Your computer is low on memory"
-- Eventually, the process failed, and this message appeared in the osql.exe window:
-- 
-- Msg 701, Level 17, State 123, Server BRECK-PC\SQLEXPRESS, Line 2
-- There is insufficient system memory in resource pool 'internal' to run this
-- query.
-- Location:        qxcntxt.cpp:1052
-- Expression:      cref == 0
-- SPID:            52
-- Process ID:      2044
-- 
-- Failed at 2009-10-12 10:22:21.817
-- The server disk I/O light remained lit without interruption.
-- The server required a hard reboot.

---------------------------------------------------------
-- Test 2 - Code as shown above, except for VARCHAR ( 8000 ).
-- SUCCESSFUL 

BEGIN
   DROP TABLE sa_target;
   EXCEPTION WHEN OTHERS THEN
END;

CREATE TABLE sa_target (
   primary_key      INTEGER NOT NULL PRIMARY KEY,
   string_column_1  VARCHAR ( 8000 ) NOT NULL );

-- Started at 2009-10-12 10:41:46.427
-- There was some slight initial growth in RAM usage on the server.
-- Server RAM usage stabilized at 1.40 GB, Physical Memory: 35%, Page File 1560M / 8352M
-- The target insert rate remained about 2000 rows per second throughout.
-- Finished OK at 2009-10-12 10:50:52.240

---------------------------------------------------------
-- Test 3 - Repeat Test 2
-- SUCCESSFUL 

-- Started at 2009-10-12 10:53:38.350
-- No further RAM usage growth on the server.
-- Finished OK at 2009-10-12 11:02:10.457

答案 2 :(得分:0)

您可以尝试一次批量运行插入而不是整个数据集。

答案 3 :(得分:0)

您可以尝试使用变量表吗?

而不是使用临时表

例如。

DECLARE @ResultTable TABLE (TableId INT PRIMARY KEY, ... etc)

INSERT INTO @ResultTable
SELECT TableId, ....
FROM mss_t2 

... etc. ...

答案 4 :(得分:0)

我遇到了类似的问题,我的代码在循环中包含一个简单的#temp表的使用,导致了这个问题,我用永久表替换了。

似乎正在发挥作用。

由于 纳温