SQL server memory management on cloud

Preface ： as everyone knows ,cpu, Memory , Disk is one of the three core resources of a service , This chapter will introduce SQL Server Internal memory structure and memory management . Finally, the memory in Tencent cloud SQL Server Reactions in cloud database monitoring indicators , Help users understand SQL Server Cloud database features .

1. SQL Server Database and NUMA

NUMA（ Inconsistent memory architecture ） To expand the performance of the motherboard . In more than one CPU Server , To avoid the shared memory bus becoming a bottleneck ,NUMA For each CPU Both have separate memory controllers .SQL Server Our infrastructure is based on NUMA The design of the . If you have known SQL Server Kernel task call part , Then you won't be right node,scheduler, Wait for the name to have a certain understanding . When SQL Server When the instance starts , Will be based on the detected NUMA Create the corresponding node node , Then according to the detected node The number of logical processors in the create the corresponding scheduler.

SELECT * FROM sys.dm_os_nodes;

see SQL Server After starting node Number of nodes .

SELECT * FROM sys.dm_os_scheduler;

see SQL Server After starting scheduler The number of , And the node The relationship between .

2. SQL Server Of memory

1. Memory nodes

Memory nodes are just as mentioned above , and NUMA Corresponding , One SQL Server The instance will have at least one memory node during startup . You can view the memory nodes with the following command .

select * from sys.dm_os_memory_nodes;

memory_mode_id and NUMA Of node The node corresponding to the

2. Memory allocator

All components require memory allocators to access memory nodes to allocate memory , Compare memory consuming components with dedicated memory allocators , You can use the following statement to view the memory member .

select type,memory_node_id,virtual_memory_reserved_kb,virtual_memory_committed_kb from sys.dm_os_memory_clerks order by virtual_memory_reserved_kb desc;

The largest memory allocator is MEMORYCLERK_SQLBIFFERPOLL, The object that allocates memory for the buffer pool , Memory node belonging to

3. cache

A lot of useful data will be cached in memory , For example, execute plan caching , Data caching , stored procedure , View caching, etc . The largest one is the buffer pool , This is where the user data is cached . Use the following SQL You can query the size of the buffer pool used by each database .

select COUNT(*)*8/1024 AS 'Catch Size (MB)',case database_id when 32767 then 'resourceDB' else db_name(database_id) end as 'Dtatbases'  from sys.dm_os_buffer_descriptors group by DB_NAME(database_id),database_id order by 'Catch Size (MB)' desc;

The database cache size of different databases in the buffer pool

3. How to determine SQL Server The cloud instance is out of memory .

Usually ：

select counter_name, ltrim(cntr_value*1.0/1024.0) as MB from master.sys.dm_os_performance_counters where counter_name like '%total_server_memory%'

View the actual total size of the current buffer pool .

select counter_name, ltrim(cntr_value*1.0/1024.0) as MB from master.sys.dm_os_performance_counters where counter_name like '%target_server_memory%'

View the desired size of the current buffer pool . It is generally considered that the ideal buffer pool size is consistent with the actual buffer pool size . In case of inconsistency, there are two situations ,1. The pressure of the business is not enough because the data is not cached .2. Not enough physical memory , You can no longer put user data in the buffer pool .

You can also use the following SQL Check in sqlserver How long a page is expected to remain in the buffer pool , The unit is seconds

select counter_name, cntr_value as s from master.sys.dm_os_performance_counters where counter_name like '%page_life_expectancy%'

When memory is tight , Pages in the buffer pool will be swapped out of memory , So the retention time will be shorter . If you keep time in hell 300 Second, you need to pay attention to the memory condition .

SQL Server Cloud instances provide rich monitoring indicators , for example ：“ Buffer cache hit rate ”,“ Execution plan cache hit rate ”,“ Memory footprint ”,“ Maximum memory ”. From these indicators, we can judge whether the instance memory is insufficient .

The console provides “ Memory footprint ” The corresponding is Total Server Memory (KB) indicators .

The console provides “ Maximum memory ” The corresponding is Target Server Memory (KB) indicators .

The console provides “ Buffer hit rate ” The corresponding is Buffer cache hit ratio indicators .

The console provides “ Execution plan cache hit rate ” The corresponding is Cache Hit Ratio indicators .

“ Memory footprint “ This indicator , After the instance is started, it runs for a period of time , It will be basically consistent with the maximum memory of the instance purchased on the cloud , Don't worry too much . You can focus on “ Buffer hit rate ” indicators , If the hit rate continues to be lower than 90%, Consider whether you need to increase memory .