诺禾，诺禾sqlserver日常问题

Posted 2020-12-23

上一篇主要介绍了sqlserver巡检时所作的数据库、数据文件、日志文件及I/O统计信息的巡检，今天主要介绍一下对象的一些检查以及性能方面的一些巡检，仅供参考。
一、对象检查

对象检查包括：表、索引、索引视图等

1、查看某个表的行数,保留大小，数据大小，索引大小，未使用大小

--不一定准确:无性能影响
exec sp_spaceused @objname =‘temp_lock‘
--准确:但有性能影响
exec sp_spaceused @objname =‘temp_lock‘,
@updateusage =‘true‘

2、一次计算多个对象的空间使用情况 （按页统计，没有性能影响，有时不准确）

sys.dm_db_partition_stats返回当前数据库中每个分区(表和索引)的页和行计数信息

select o.name,
sum(p.reserved_page_count) as reserved_page_count, --保留页，包含表和索引
sum(p.used_page_count) as used_page_count, --已使用页，包含表和索引
sum(case when p.index_id <2
then p.in_row_data_page_count +
p.lob_used_page_count +
p.row_overflow_used_page_count
else p.lob_used_page_count +
p.row_overflow_used_page_count
end) as data_pages, --数据页,包含表中数据、索引中的lob数据、索引中的行溢出数据
sum(case when p.index_id < 2
then p.row_count
else 0
end) as row_counts --数据行数，包含表中的数据行数，不包含索引中的数据条目数
from sys.dm_db_partition_stats p
inner join sys.objects o
on p.object_id = o.object_id
--where p.object_id= object_id(‘表名‘)
group by o.name

sqlserver数据库日常巡检规范之下篇，值得收藏

3、显示当前数据库中所有的表或视图的数据和索引的空间信息（按页或区统计，有性能影响，准确）

这里包含：逻辑碎片、区碎片(碎片率)、平均页密度

insert extentinfo exec(‘dbcc extentinfo(‘‘test‘‘) ‘) go

ps:
sqlserver数据库日常巡检规范之下篇，值得收藏

4、查找数据库需重建的对象

4.1、建表

use dbname
go
if OBJECT_ID(‘extentinfo‘) is not null
drop table extentinfo
go

create table extentinfo
( [file_id] smallint,
page_id int,
pg_alloc int,
ext_size int,
obj_id int,
index_id int,
partition_number int,
partition_id bigint,
iam_chain_type varchar(50),
pfs_bytes varbinary(10)
)
go

4.2、查询数据文件盘区信息

查询到的盘区信息是数据库的数据文件的盘区信息，日志文件不以盘区为单位
命令格式: DBCC EXTENTINFO（dbname,tablename,indexid）
DBCC EXTENTINFO(‘[test]‘,‘extentinfo‘,0)

insert extentinfo exec(‘dbcc extentinfo(‘‘test‘‘) ‘) go

4.3、每一个区有一条数据

select file_id,
obj_id, --对象ID
index_id, --索引id
page_id, --这个区是从哪个页开始的,也就是这个区中的第一个页面的页面号
pg_alloc, --这个盘区分配的页面数量
ext_size, --这个盘区包含了多少页
partition_number,
partition_id,
iam_chain_type, --IAM链类型:行内数据,行溢出数据,大对象数据
pfs_bytes
from extentinfo
order by file_id,
OBJ_ID,
index_id,
partition_id,
ext_size

4.4、重建对象
数据库的数据文件的盘区信息,通过计算每个对象理论上区的数量和实际数量,如果两者相差很大,
那就应该重建对象.
1)每一条记录就是一个区
2)如果pg_alloc比ext_size小，也就是实际每个区分配的页数小于理论上这个区的页数,那么就会多一条记录，把本应该属于这个区的页放到多出来的这条记录

对应的区中,那么原来只有一条记录(也就是一个区)，现在就有2条记录(也就是2个区),导致实际的区数量2大于理论上的区数量1.

select file_id,
obj_id,
index_id,
partition_id,
ext_size,
count() as ‘实际区的个数‘,
sum(pg_alloc) as ‘实际包含的页数‘,
ceiling(sum(pg_alloc) 1.0 / ext_size) as ‘理论上的区的个数‘,
ceiling(sum(pg_alloc) 1.0 / ext_size) / count() 100.00 as ‘理论上的区个数 / 实际区的个数‘
from extentinfo
group by file_id,
obj_id,
index_id,
partition_id,
ext_size
having ceiling(sum(pg_alloc)1.0/ext_size) < count(*)
--过滤: 理论上区的个数 < 实际区的个数,也就是百分比小于100%的
order by partition_id, obj_id, index_id, [file_id]

六、tempdb数据库检查

监控tempdb的空间使用方法:

（1）可以通过SQL Trace来跟踪,但是由于不能预期造成大量使用tempdb语句在什么时候运行, 而且SQL Trance操作比较昂贵,如果一直开着会产生大量的跟踪文件,对硬盘的负担也比较重,一般不用.

（2）轻量级的监控是通过一定时间间隔运行能够监控系统运行的dbcc命令、动态性能视图-函数， 把结果记录在文件中，这对于很繁忙的系统是不错的选择。

1、tempdb概览

SELECT t1.session_id, t1.internal_objects_alloc_page_count, t1.user_objects_alloc_page_count, t1.internal_objects_dealloc_page_count , t1.user_objects_dealloc_page_count, t.text from sys.dm_db_session_space_usage t1 --反映每个session的累计空间申请 inner join sys.dm_exec_sessions as t2 on t1.session_id = t2.session_id inner join sys.dm_exec_requests t3 on t2.session_id = t3.session_id cross apply sys.dm_exec_sql_text(t3.sql_handle) t where t1.internal_objects_alloc_page_count>0 or t1.user_objects_alloc_page_count >0 or t1.internal_objects_dealloc_page_count>0 or t1.user_objects_dealloc_page_count>0

sqlserver数据库日常巡检规范之下篇，值得收藏

6.2、查看tempdb空间的总体分配,申请空间的会话正在运行的语句

SELECT
t1.session_id,
t1.internal_objects_alloc_page_count,
t1.user_objects_alloc_page_count,
t1.internal_objects_dealloc_page_count ,
t1.user_objects_dealloc_page_count,
t.text
from sys.dm_db_session_space_usage t1 --反映每个session的累计空间申请
inner join sys.dm_exec_sessions as t2
on t1.session_id = t2.session_id
inner join sys.dm_exec_requests t3
on t2.session_id = t3.session_id
cross apply sys.dm_exec_sql_text(t3.sql_handle) t
where t1.internal_objects_alloc_page_count>0 or
t1.user_objects_alloc_page_count >0 or
t1.internal_objects_dealloc_page_count>0 or
t1.user_objects_dealloc_page_count>0

sqlserver数据库日常巡检规范之下篇，值得收藏

6.3、返回tempdb中页分配和释放活动，
只有当任务正在运行时，
sys.dm_db_task_space_usage才会返回值。在请求完成时，这些值将按session聚合体现在
SYS.dm_db_session_space_usage

select t.session_id,
t.request_id,
t.database_id,
t.user_objects_alloc_page_count,
t.internal_objects_dealloc_page_count,
t.internal_objects_alloc_page_count,
t.internal_objects_dealloc_page_count
from sys.dm_db_task_space_usage t
inner join sys.dm_exec_sessions e
on t.session_id = e.session_id
inner join sys.dm_exec_requests r
on t.session_id = r.session_id and
t.request_id = r.request_id

sqlserver数据库日常巡检规范之下篇，值得收藏

七、性能相关

1、查询SQL Server执行过的SQL语句（执行次数）

有的时候，需要知道近段时间SQLSERVER执行了什么语句，可以用下面的方法：

SELECT ss.SUM_execution_count , t.text , ss.SUM_total_elapsed_time , ss.sum_total_worker_time , ss.sum_total_logical_reads , ss.sum_total_logical_writes FROM ( SELECT S.plan_handle , SUM(s.execution_count) SUM_Execution_count , SUM(s.total_elapsed_time) SUM_total_elapsed_time , SUM(s.total_worker_time) SUM_total_worker_time , SUM(s.total_logical_reads) SUM_total_logical_reads , SUM(s.total_logical_writes) SUM_total_logical_writes FROM sys.dm_exec_query_stats s GROUP BY S.plan_handle ) AS ss CROSS APPLY sys.dm_exec_sql_text(ss.plan_handle) t ORDER BY sum_total_logical_reads DESC

sqlserver数据库日常巡检规范之下篇，值得收藏

2、开销较大的查询

SELECT ss.SUM_execution_count ,
t.text ,
ss.SUM_total_elapsed_time ,
ss.sum_total_worker_time ,
ss.sum_total_logical_reads ,
ss.sum_total_logical_writes
FROM ( SELECT S.plan_handle ,
SUM(s.execution_count) SUM_Execution_count ,
SUM(s.total_elapsed_time) SUM_total_elapsed_time ,
SUM(s.total_worker_time) SUM_total_worker_time ,
SUM(s.total_logical_reads) SUM_total_logical_reads ,
SUM(s.total_logical_writes) SUM_total_logical_writes
FROM sys.dm_exec_query_stats s
GROUP BY S.plan_handle
) AS ss
CROSS APPLY sys.dm_exec_sql_text(ss.plan_handle) t
ORDER BY sum_total_logical_reads DESC

sqlserver数据库日常巡检规范之下篇，值得收藏

3、找出执行时间最长的10条SQL（适用于SQL SERVER 2005及其以上版本）

如果想对SQL作筛选，可将not like ‘%fetch%‘ 换成 like ‘%user%‘就可以找出SQL语句中含有user关键字的SQL

SELECT (total_elapsed_time / execution_count)/1000 N‘平均时间ms‘ ,total_elapsed_time/1000 N‘总花费时间ms‘ ,total_worker_time/1000 N‘所用的CPU总时间ms‘ ,total_physical_reads N‘物理读取总次数‘ ,total_logical_reads/execution_count N‘每次逻辑读次数‘ ,total_logical_reads N‘逻辑读取总次数‘ ,total_logical_writes N‘逻辑写入总次数‘ ,execution_count N‘执行次数‘ ,SUBSTRING(st.text, (qs.statement_start_offset/2) + 1 ,((CASE statement_end_offset WHEN -1 THEN DATALENGTH(st.text) ELSE qs.statement_end_offset END - qs.statement_start_offset)/2) + 1) N‘执行语句‘ ,creation_time N‘语句编译时间‘ ,last_execution_time N‘上次执行时间‘ FROM sys.dm_exec_query_stats AS qs CROSS APPLY sys.dm_exec_sql_text(qs.sql_handle) st WHERE SUBSTRING(st.text, (qs.statement_start_offset/2) + 1, ((CASE statement_end_offset WHEN -1 THEN DATALENGTH(st.text) ELSE qs.statement_end_offset END - qs.statement_start_offset)/2) + 1) not like ‘fetch%‘ ORDER BY total_elapsed_time / execution_count DESC;

sqlserver数据库日常巡检规范之下篇，值得收藏

4、找出执行最慢的SQL语句（适用于SQL SERVER 2005及其以上版本）

--给N赋初值为30 declare @n int set @n=30;with maco as ( select top (@n) plan_handle, sum(total_worker_time) as total_worker_time , sum(execution_count) as execution_count , count(1) as sql_count from sys.dm_exec_query_stats group by plan_handle order by sum(total_worker_time) desc ) select t.text , a.total_worker_time , a.execution_count , a.sql_count from maco a cross apply sys.dm_exec_sql_text(plan_handle) t

sqlserver数据库日常巡检规范之下篇，值得收藏

5、找出最耗时的前N条T-SQL语句 （适用于SQL SERVER 2005及其以上版本）

SELECT TOP 20 total_worker_time/1000 AS [总消耗CPU 时间(ms)],execution_count [运行次数], qs.total_worker_time/qs.execution_count/1000 AS [平均消耗CPU 时间(ms)], last_execution_time AS [最后一次执行时间],min_worker_time /1000 AS [最小执行时间(ms)], max_worker_time /1000 AS [最大执行时间(ms)], SUBSTRING(qt.text,qs.statement_start_offset/2+1, (CASE WHEN qs.statement_end_offset = -1 THEN DATALENGTH(qt.text) ELSE qs.statement_end_offset END -qs.statement_start_offset)/2 + 1) AS [使用CPU的语法], qt.text [完整语法], qt.dbid, dbname=db_name(qt.dbid), qt.objectid,object_name(qt.objectid,qt.dbid) ObjectName FROM sys.dm_exec_query_stats qs WITH(nolock) CROSS apply sys.dm_exec_sql_text(qs.sql_handle) AS qt WHERE execution_count>1 ORDER BY (qs.total_worker_time/qs.execution_count/1000) DESC

6、平均耗CPU最多的前20个SQL （SQL SERVER 2008或以上版本）

SELECT TOP 20 total_worker_time/1000 AS [总消耗CPU 时间(ms)],execution_count [运行次数], qs.total_worker_time/qs.execution_count/1000 AS [平均消耗CPU 时间(ms)], last_execution_time AS [最后一次执行时间],max_worker_time /1000 AS [最大执行时间(ms)], SUBSTRING(qt.text,qs.statement_start_offset/2+1, (CASE WHEN qs.statement_end_offset = -1 THEN DATALENGTH(qt.text) ELSE qs.statement_end_offset END -qs.statement_start_offset)/2 + 1) AS [使用CPU的语法], qt.text [完整语法], qt.dbid, dbname=db_name(qt.dbid), qt.objectid,object_name(qt.objectid,qt.dbid) ObjectName FROM sys.dm_exec_query_stats qs WITH(nolock) CROSS apply sys.dm_exec_sql_text(qs.sql_handle) AS qt WHERE execution_count>1 ORDER BY total_worker_time DESC

7、总耗CPU最多的前20个SQL （SQL SERVER 2008或以上版本）

SELECT TOP 20
total_worker_time/1000 AS [总消耗CPU 时间(ms)],execution_count [运行次数],
qs.total_worker_time/qs.execution_count/1000 AS [平均消耗CPU 时间(ms)],
last_execution_time AS [最后一次执行时间],max_worker_time /1000 AS [最大执行时间(ms)],
SUBSTRING(qt.text,qs.statement_start_offset/2+1,
(CASE WHEN qs.statement_end_offset = -1
THEN DATALENGTH(qt.text)
ELSE qs.statement_end_offset END -qs.statement_start_offset)/2 + 1)
AS [使用CPU的语法], qt.text [完整语法],
qt.dbid, dbname=db_name(qt.dbid),
qt.objectid,object_name(qt.objectid,qt.dbid) ObjectName
FROM sys.dm_exec_query_stats qs WITH(nolock)
CROSS apply sys.dm_exec_sql_text(qs.sql_handle) AS qt
WHERE execution_count>1
ORDER BY total_worker_time DESC

8、逻辑读最高的前25条语句

select top 25
p.name as [SP Name],
deps.total_logical_reads as [TotalLogicalReads],
deps.total_logical_reads / deps.execution_count as [AvgLogicalReads],
deps.execution_count,
ISNULL(deps.execution_count/DATEDIFF(Second,deps.cached_time,Getdate()),0) as [Calls/Second],
deps.total_elapsed_time,
deps.total_elapsed_time / deps.execution_count as [avg_elapsed_time] ,
deps.cached_time
from sys.procedures as p
inner join sys.dm_exec_procedure_stats as deps on p.[object_id] = deps.[object_id]
where deps.database_id = DB_ID()
order by deps.total_logical_reads desc

9、最耗时的前100个存储过程

SELECT TOP
100 a.object_id,--对象名称
a.database_id,--数据库名称
OBJECT_NAME( object_id, database_id ) AS ‘proc name‘,--存储过程名称
a.cached_time,--存储过程添加到缓存的时间
a.last_execution_time,--上次执行时间
a.total_elapsed_time,--总耗时
a.total_elapsed_time/ a.execution_count AS [avg_elapsed_time],--平均耗时
a.execution_count,--运行次数
a.total_physical_reads/ a.execution_count avg_physical_reads,--平均物理读取数
a.total_logical_writes,--总逻辑写入次数
a.total_logical_writes/ a.execution_count avg_logical_reads,--平均逻辑写入次数
a.last_elapsed_time,--上次运行耗时
a.[total_worker_time] --此存储过程自编译以来执行所用的 CPU 时间总量（微秒）
FROM
sys.dm_exec_procedure_stats a
ORDER BY
[avg_elapsed_time] DESC;

sqlserver数据库日常巡检规范之下篇，值得收藏

八、阻塞锁及长任务检查

1、死锁等待

SELECT
wt.session_id AS waiting_session_id,
DB_NAME( tl.resource_database_id ) AS DatabaseName,
wt.wait_duration_ms
,
wt.waiting_task_address,
tl.request_mode
,
(
SELECT SUBSTRING
(
st.text,
( er.statement_start_offset/ 2 ) + 1,
( CASE er.statement_end_offset WHEN - 1 THEN datalength( st.text ) ELSE er.statement_end_offset END - er.statement_start_offset/ 2 ) + 1
)
FROM
sys.dm_exec_requests AS er CROSS apply sys.dm_exec_sql_text ( er.sql_handle ) AS st
WHERE
er.session_id= tl.request_session_id
) AS waiting_query_text,
tl.resource_type,
tl.resource_associated_entity_id,
wt.wait_type,
wt.blocking_session_id
,
wt.resource_description AS blocking_resource_description,
CASE

    WHEN wt.blocking_session_id> 0 THEN
    (
    SELECT
        st2.text 
    FROM
        sys.sysprocesses AS sp CROSS apply sys.dm_exec_sql_text ( sp.sql_handle ) AS st2 
    WHERE
        sp.spid= wt.blocking_session_id 
    ) ELSE NULL 
END AS blocking_query_text 

FROM
sys.dm_os_waiting_tasks AS wt
JOIN sys.dm_tran_locks AS tl ON wt.resource_address = tl.lock_owner_address
WHERE
wt.wait_duration_ms > 5000;

sqlserver数据库日常巡检规范之下篇，值得收藏

2、表锁跟踪

select request_session_id,resource_type, DB_NAME(resource_database_id) as DatabaseName
,OBJECT_NAME(resource_associated_entity_id) as TableName, request_mode, request_type, request_status
from sys.dm_tran_locks as tl join sys.all_objects as ao on tl.resource_associated_entity_id=ao.object_id
where request_type=‘LOCK‘ and request_status=‘GRANT‘ and request_mode in(‘X‘,‘S‘)
and ao.type=‘U‘ and resource_type=‘OBJECT‘ and tl.resource_database_id=db_id()

3、长时间等待的任务

select top 20 wt.session_id, wt.waiting_task_address, wt.wait_duration_ms, wt.wait_type
, wt.blocking_session_id, wt.resource_description
--, wt.waiting_tasks_count, wt.max_wait_time_ms
from sys.dm_os_waiting_tasks as wt
-- where wt.wait_duration_ms > 5000
order by wait_duration_ms desc

sqlserver数据库日常巡检规范之下篇，值得收藏

4、Top 20 耗时任务

SELECT TOP 20 creation_time, last_execution_time, total_elapsed_time,
total_elapsed_time/execution_count AS [Avg Elapsed Time us], last_elapsed_time,
execution_count
, ( SELECT SUBSTRING(text, statement_start_offset/2,
( CASE
WHEN statement_end_offset = -1
THEN LEN(CONVERT(nvarchar(max),text)) * 2
ELSE statement_end_offset END

• statement_start_offset
  )
  )
  FROM sys.dm_exec_sql_text(sql_handle)) AS query_text
  FROM sys.dm_exec_query_stats
  ORDER BY [Avg Elapsed Time us] DESC;

sqlserver数据库日常巡检规范之下篇，值得收藏

觉得有用的朋友多帮忙转发哦！后面会分享更多devops和DBA方面的内容，感兴趣的朋友可以关注下~