In-Memory：内存优化表 DMV

SQL Server 在执行查询时，自动将活动的相关信息保存在内存中，这些活动信息称作DMV（Dynamic Management View），DMV记录SQL Server实例级别上的活动信息。由于DMV使用内存作为存储媒介，在读取DMV时，不需要IO操作，读写数据速度极快，不会对Server产生压力，并且DMV直接存储在服务器的内存中，能够及时、精确地反映系统性能的最新状态。

一，使用DMV的注意事项

1，确定数据保存的时间

内存是易失性的存储媒介，一旦SQL Server实例重启，DMV存储的信息将全部重置。在使用DMV时，首先需要检查这些信息在内存中保存了多长时间，以确定DMV数据的可用性。如果SQL Server仅仅运行很短的一段时间，那么对DMV数据进行统计和分析是不合适的，这些数据不是SQL Server 实例真实工作负载的数据样本。SQL Server运行的时候越长，DMV中保存的信息就越多（当然，DMV非常小，不会对内存造成压力），利用DMV分析就越准确。

2，DMV使用的内存有限

DMV能够使用的内存容量有限，这使得DMV只能存储有限数量的数据。如果SQL Server运行了很长世间，SQL Server Engine会将DMV的一些老数据覆盖。

二，查看内存优化表的DMV

SQL Server创建一些DMV，用于追踪内存优化表的活动信息，在内存优化表的DMV中，有两个对象ID（Object ID）：

• xtp_object_id 是内部的内存优化表（Internal Memory-Optimized Table)的ID，在对象的整个生命周期中，该ID可变；
• object_id 是User Table的ID，唯一标识该User Table，在对象的整个生命周期中，该ID不变；

xtp_object_id 是内部的内存优化表的ID（Internal Memory-Optimized Table)，每一个User Table都对应一个或多个Internal Table：

• 一个Interal Table用于存储核心数据（Core Data）；
• 其他的Internal Table 用于存储临时数据，Columnstore Index ，Off-Row（Lob）；
• 详细信息，请阅读DMV: sys.memory_optimized_tables_internal_attributes

1，查看内存优化表占用的物理内存

--memory usage
select tms.object_id
    ,object_schema_name(tms.object_id)+'.'+object_name(tms.object_id) as table_name
    ,(tms.memory_allocated_for_indexes_kb+tms.memory_allocated_for_table_kb)/1024 as total_allocated_mb
    ,tms.memory_allocated_for_table_kb/1024 as table_allocated_mb
    ,tms.memory_used_by_table_kb/1024 as table_used_mb
    ,(tms.memory_allocated_for_table_kb-tms.memory_used_by_table_kb)/1024 as table_unused_mb
    ,tms.memory_allocated_for_indexes_kb/1024 as index_allocated_mb
    ,tms.memory_used_by_indexes_kb/1024 as index_used_mb
    ,(tms.memory_allocated_for_indexes_kb-tms.memory_used_by_indexes_kb)/1024 as index_unused_mb
from sys.dm_db_xtp_table_memory_stats tms
where tms.object_id>0

2，查看内存消费者（Memory Consumer）

每一个MOT都有单独的Memory Heap，称作VarHeap，是一个Memory Consumer，SQL Server从VarHeap中为MOT的数据分配内存空间。varheap是可变大小的堆数据结构，能够收缩和增长。VarHeap是由固定数量的Allocation Unit组成的集合。Allocation Unit用于分配特定大小的Page，Page的大小是不固定的，最常见的Page Size是64KB。

VarHeap用于Table Row 和 Bw-Tree Index。每一个LOB列（使用max指定大小）都有自己独立的VarHeap。在创建MOT时，SQL Server决定哪些column存储在Table的VarHeap中，哪些column存储在自己独立的VarHeap中。

Hash Index 使用另外一个Memory Consumer，称作Hash。

select
    object_schema_name(mc.object_id)+'.'+object_name(mc.object_id) as table_name
    ,a.xtp_object_id
    ,a.type_desc as xtp_object_type
    ,iif(a.minor_id=0,'User Table','Off-Row Column:'+col_name(a.object_id,a.minor_id)) as xtp_object
    ,mc.memory_consumer_id as consumer_id
    ,mc.memory_consumer_type_desc as consumer_type
    ,mc.memory_consumer_desc as consumer_desc
    ,i.name as index_name
    ,i.type_desc as index_type_desc
    ,mc.allocated_bytes/1024/1024 as allocated_mb
    ,mc.used_bytes/1024/1024 as used_mb
    ,mc.allocation_count
from sys.dm_db_xtp_memory_consumers mc
inner join sys.memory_optimized_tables_internal_attributes a
    on mc.object_id=a.object_id
        and mc.xtp_object_id=a.xtp_object_id
left join sys.indexes i
    on mc.object_id=i.object_id
        and mc.index_id=i.index_id
where mc.object_id=object_id('[influencer].[Influencers]')

引用：SQL Server In-Memory OLTP Internals for SQL Server 2016

The varheaps are used for both table rows and Bw-tree indexes. (Hash indexes are the only structure used with memory-optimized tables that uses a different memory consumer.) In addition, each LOB, column (specified with the MAX qualifier in the datatype definition) has its own varheap. As mentioned earlier, SQL Server 2016 also supports large columns similar to the row-overflow columns for disk-based tables. For memory-optimized tables, SQL Server will decide when the table is created which of your variable length columns will be stored in the table’s varheap and which will be stored as overflow data and have their own varheap. You can think of LOB and row-overflow columns as being stored in their own internal tables.

You can examine the metadata for each varheap (and the other memory consumers) in a DMV called sys.dm_db_xtp_memory_consumers. Each memory-optimized table has a row in this view for each varheap and for each hash index. (We will see one more type of memory consumer, called HkCS Allocator, in the section on columnstore indexes on memory-optimized tables.) If we join this DMV with the catalog view called sys.memory_optimized_tables_internal_attributes we can see which varheap belongs to a specific column. Each user table has a unique object_id and xtp_object_id value which is used by all the indexes. Each additional varheap, for the row_overflow and LOB columns will have its own xtp_object_id. Note that if an object is altered, its xtp_object_id will change, but its object_id will not.

3，Hash Index的链表长度

对于Hash Index，表示链长的字段有：avg_chain_length 和 max_chain_length ，链长应保持在2左右；链长过大，表明太多的数据行被映射到相同的Bucket中，这会显著影响DML操作的性能。

导致链长过大的原因是：

• 总的Bucket数量少，导致不同的Index Key映射到相同的Bucket上；
• 如果空的Bucket数量大，但链长过大，这说明，Hash Index存在大量重复的Index Key；相同的Index Key被映射到相同的bucket；
• 详细信息，请阅读：sys.dm_db_xtp_hash_index_stats (Transact-SQL)；

4，事务

在访问MOT时，有两种类型事务ID，在事务中，访问MOT和访问DBT的事务是独立的：

• xtp_transaction_id：是访问MOT的事务ID；
• transaction_id：是访问DBT的事务ID，0表示事务没有访问MOT；
• 详细信息，请阅读：sys.dm_db_xtp_transactions (Transact-SQL)

查看当前数据库中活跃事务的信息：

select t.xtp_transaction_id
    ,t.transaction_id
    ,t.session_id
    ,t.begin_tsn
    ,t.end_tsn
    ,t.state_desc
    ,t.result_desc
from sys.dm_db_xtp_transactions t

参考文档：

Baselining with SQL Server Dynamic Management Views

Memory-Optimized Table Dynamic Management Views (Transact-SQL)