One of the topics covered during IE2 is Waits and Queues, and this week one of the attendees asked if there was a way to know when wait stats were last cleared. It turns out there is a post about it on the SQLOS & Cloud Infrastructure Team Blog, and what you do is look for the SQLTRACE_INCREMENTAL_FLUSH_SLEEP wait_type in sys.dm_os_wait_stats. The value for wait_tims_ms is the time since wait stats were last cleared.
Since I talk about clearing wait stats in my DBCC session I decided this information would be useful to include, and then decided to just write a query to figure it out. The query is below, as is a query to show the last time the instance was restarted. Edit: Note that this query is only for SQL Server 2008R2 and higher. The SQLTRACE_INCREMENTAL_FLUSH_SLEEP wait_type did not exist in previous versions of SQL Server. If I can find an alternate wait_type for use, I will post back.
Just to be clear, I am not advocating that you immediately clear wait stats on your server. In fact, I’d be more interested to know whether wait stats have been cleared since the instance was last restarted. Let me back up a bit… If you’re not familiar with wait statistics, I suggest starting with Paul’s post, Wait statistics, or please tell me where it hurts. Whether you have performance issues on your server or not, I highly recommend looking at wait stats regularly to understand what your “normal” waits are. If you are having issues, waits can be a starting point to sorting out what’s causing the problem. But wait stats should not be viewed in a vacuum – always use other counters or information to correlate findings.
Once you have a basic understanding of your wait stats, you may decide to capture them on a regular basis. But wait stats are cumulative, and will only reset when you restart the instance, or if you clear them manually. They can be cleared using:
DBCC SQLPERF (‘sys.dm_os_wait_stats’, CLEAR);
There are a variety of reasons to clear wait stats, but it shouldn’t be done arbitrarily. You may clear them because you’ve made a change to the system and want to understand if and how that change affects wait statistics. You may clear wait stats every morning at 5 AM because backups and other maintenance tasks run at night, but you want to understand performance during business hours when users are the system and therefore decide to filter out (or investigate separately) waits from those tasks. In the latter case, you probably have a scheduled job to clear the wait stats, and other DBAs may or may not know about this job. If another DBA decides to clear wait stats for some reason, without telling you, they may affect the wait information you’ve been capturing. As such, I would always verify with colleagues whether or not wait stats are cleared regularly and this script will give you an idea of whether are being cleared or not. However, even if they haven’t been cleared since the last restart, I would still perform some due diligence and verify it won’t alter anyone else’s data.
I am interested to hear different ideas for how this information could be used, so feel free to post back or email me. And if you’re not currently looking at your wait stats regularly, please start!
/* when were wait stats last cleared? */ SELECT wait_type, wait_time_ms, DATEADD(ms,-wait_time_ms,getdate()) as “Date/TimeCleared”, CASE WHEN wait_time_ms < 1000 THEN cast(wait_time_ms as varchar(15)) + ‘ ms’ WHEN wait_time_ms between 1000 and 60000 THEN cast((wait_time_ms/1000) as varchar(15)) + ‘ seconds’ WHEN wait_time_ms between 60001 and 3600000 THEN cast((wait_time_ms/60000) as varchar(15)) + ‘ minutes’ WHEN wait_time_ms between 3600001 and 86400000 THEN cast((wait_time_ms/3600000) as varchar(15)) + ‘ hours’ WHEN wait_time_ms > 86400000 THEN cast((wait_time_ms/86400000) as varchar(15)) + ‘ days’ END as “TimeSinceCleared” FROM sys.dm_os_wait_stats WHERE wait_type = ‘SQLTRACE_INCREMENTAL_FLUSH_SLEEP’; /* check SQL Server start time – 2008 and higher */ SELECT sqlserver_start_time FROM sys.dm_os_sys_info;