Tuesday, 12 September 2017

Detecting poor cursor usage in SQL Server

Without a shadow of a doubt, poor cursor usage issue is an essential aspect to review so as to reduce the possibility of having CPU bottleneck issues, and determine whether cursors are the most appropriate means to accomplish the processing or whether a set-based operation is possible. It is well know that set-based operation is generally more efficient, but if you make the decision of using cursors, you should ensure that it does not represent an issue for the database in question. Thus, detecting poor cursor usage and taking certain measures to ease the problem is crucial, and in SQL Server there are some techniques available to be used to diagnose the issue. To begin with, by using performance counters we are able to analyse the issue and find out the extent to which poor cursor usage is adversely affecting on the performance of specific workloads or the whole database. For instance, here is the counter SQL Server: Cursor Manager By Type: Cursor Requests/Sec which retrieves information about the number of SQL cursor requests received by the server. On the top of that, it is possible to filter out by the cursor manager instance such as API Cursor (only the API cursor information), TSQL Global Cursor and TSQL Local Cursor. Have a look at SQL Server, Cursor Manager by Type Object to get more info about the counter.
On the other hand, using SQL Trace is also helpful, for example, use a trace that includes the RPC:Completed event class search for 'sp_cursorfetch' statements. The value of the fourth parameter is the number of rows returned by the fetch. It is worth noting that the maximum number of rows that are requested to be returned is specified as an input parameter in the corresponding RPC:Starting event class. Finally, by using the DMV 'sys.dm_exec_cursors' we can also determine whether poor cursor usage exists in the database server as Transact-SQL cursors always have a fetch buffer of 1 and for API cursors it should be higher.

select c.* 
from sys.dm_exec_sessions s
    cross apply sys.dm_exec_cursors(con.session_id) as c
where cur.fetch_buffer_size = 1 
    and cur.properties LIKE 'API%'

Consequently, if it is seen that API Cursors have a fetch buffer size of 1 then consider enabling multiple active results (MARS) when connecting to SQL Server and consult the appropriate documentation for your specific API to determine how to specify a higher fetch buffer size for the cursor either ODBC (SQL_ATTR_ROW_ARRAY_SIZE) or OLE DB (IRowset::GetNextRows, IRowsetLocate::GetRowsAt). After that, we can retrieve more details about the session and connection of the users associated to the harmful cursors so as to decide what to do.

select s.session_id, cn.client_net_address, s.login_name, s.status,s.client_interface_name, s.program_name, 
       c.cursor_id, c.name, c.properties, c.plan_generation_num, c.creation_time, c.is_open, c.fetch_status, 
       c.fetch_buffer_size, c.worker_time, c.reads, c.writes, c.dormant_duration
from sys.dm_exec_connections cn
inner join  sys.dm_exec_sessions s on cn.session_id = s.session_id
cross apply sys.dm_exec_cursors(s.session_id) as c
where c.fetch_buffer_size = 1 
     and c.properties LIKE 'API%'

That is all for now, thanks for reading. Let me know any remarks you may have. Stay tuned.

No comments:

Post a Comment