Retrieving Deadlock Graphs with SQL Server 2008 Extended Events

Question

Post reply

Retrieving Deadlock Graphs with SQL Server 2008 Extended Events

Jonathan Kehayias

One Orange Chip

Points: 26778
More actions
February 20, 2009 at 12:26 am

#197311

Comments posted to this topic are about the item Retrieving Deadlock Graphs with SQL Server 2008 Extended Events
Jonathan Kehayias | Principal Consultant | MCM: SQL Server 2008
My Blog | Twitter | MVP Profile
Training | Consulting | Become a SQLskills Insider
Troubleshooting SQL Server: A Guide for Accidental DBAs[/url]

Viewing 15 posts - 1 through 15 (of 83 total)

You must be logged in to reply to this topic. Login to reply

ChiragNS One Orange Chip Points: 26137 More actions · Answer 1

nice article.

Is the deadlock info persisted after server/sql service restart.

"Keep Trying"

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 2

Great question. Unfortunately, the answer is no. Since the ring_buffer target is memory resident, it does not persist the information after restart. You can however, alter the system_health session and add a asyncronous_file_target to it which will log the information to a file on disk which will survive service restarts.

icata Hall of Fame Points: 3176 More actions · Answer 3

Can you alter the default system_health default session?

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 4

Jonathan Kehayias (2/23/2009)
You can however, alter the system_health session and add a asyncronous_file_target to it which will log the information to a file on disk which will survive service restarts.

As I covered in my first response, yes you can alter the default system_health session, though beyond adding a file target, I personally wouldn't change it, I would instead create a new session to handle whatever events I am interested in looking at. You can also drop it if you don't want it to exist at all. If you mess it up, you won't be capturing the information that the CSS team may want in troubleshooting a problem which may cause delays in your support case if they request that you recreate the system_health session from script.

The script that created this session is located in the InstallPath\Microsoft SQL Server\MSSQL10.MSSQLSERVER\MSSQL\Install folder in the u_tables.sql file. It is at the very bottom of the file. So if you do play with this session and screw it up, you can fix it by running the script from that file.

icata Hall of Fame Points: 3176 More actions · Answer 5

icata

Hall of Fame

Points: 3176

February 23, 2009 at 11:57 am

#948835

okay, thanks

rama.mathanmohan SSC Veteran Points: 278 More actions · Answer 6

Hi,

you can use SQL Agent with WMI alert to capture and log deadlocks it works beautifully in SQL2005 and hope should work in SQL2008 see http://msdn.microsoft.com/en-us/library/ms186385.aspx for more details.

Rama Mathanmohan

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 7

rama.mathanmohan (2/23/2009)
Hi,
you can use SQL Agent with WMI alert to capture and log deadlocks it works beautifully in SQL2005 and hope should work in SQL2008 see http://msdn.microsoft.com/en-us/library/ms186385.aspx for more details.
Rama Mathanmohan

WMI alerts are still there, but you have to turn them on, and they have a considerable cost when compared with Extended Events and the system_health session which is running by default. The cost to fire an event in Extended Events is 2 thousands of a millisecond (2us) which is insignificant when compared with the 2-3% CPU cost for the WMI deadlock monitor. The difference is that the event is buffered to the ring_buffer target asynchronously as the event fires instead of having to be read from the WMI Eventing and then processed by triggering a job. There is also consider problems with setting up WMI alerts in some environments for varying reasons. You can search the Administering forum on here and find a number of people having WMI namespace problems configuring the alerts like the following:

"The @wmi_query could not be executed in the @wmi_namespace provided. Verify that an event class selected in the query exists in the namespace and that the query has the correct syntax."

Extended Events are the future of troubleshooting SQL Server, so if you are on SQL Server 2008, I'd advise to utilize it over an older construct like WMI Alerts, Trace Flags, or Profiler Traces, all of which I have written about previously. They were all good methods of doing things in SQL 2005, but they don't compare to Extended Events in SQL 2008.

ChiragNS One Orange Chip Points: 26137 More actions · Answer 8

Jonathan Kehayias (2/23/2009)
Great question. Unfortunately, the answer is no. Since the ring_buffer target is memory resident, it does not persist the information after restart. You can however, alter the system_health session and add a asyncronous_file_target to it which will log the information to a file on disk which will survive service restarts.

Thankyou

"Keep Trying"

icata Hall of Fame Points: 3176 More actions · Answer 9

I have and had quite a few issues with SQL Server integration with WMI especially with the Configuration Tools, so I am at the point that I think that regardless what might be the issues, the implementation of WMI is just unstable. To use WMI for anything would be a hard sale for me.

Maxer SSCrazy Eights Points: 8964 More actions · Answer 10

When I run your query on my server.... it never finishes! 🙂

I have let it run for up to 13 minutes... still nothing.

----------

select CAST(

REPLACE(

REPLACE(XEventData.XEvent.value('(data/value)[1]', 'varchar(max)'),

'<victim-list>', '<deadlock><victim-list>'),

'<process-list>','</victim-list><process-list>')

as xml) as DeadlockGraph

FROM

(select CAST(target_data as xml) as TargetData

from sys.dm_xe_session_targets st

join sys.dm_xe_sessions s on s.address = st.event_session_address

where name = 'system_health') AS Data

CROSS APPLY TargetData.nodes ('//RingBufferTarget/event') AS XEventData (XEvent)

where XEventData.XEvent.value('@name', 'varchar(4000)') = 'xml_deadlock_report'

-----------

When I run it against my dev server, it completes quickly, but with no results.

Am I missing something obvious here or some additional step I need to attend to?

Thank you

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 11

You probably have a lot of events in the ring_buffer target that is causing the XML parsing to be slow. What is the output of SELECT @@VERSION on the server, and how big is the XML document in the target_data?

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 12

Looking at the XQuery code a year later, I do see that there are some things that are not necessarily written well. Try this one out:

select CAST(

REPLACE(

REPLACE(XEventData.XEvent.value('(data/value)[1]', 'varchar(max)'),

'<victim-list>', '<deadlock><victim-list>'),

'<process-list>','</victim-list><process-list>')

as xml) as DeadlockGraph

FROM

(select CAST(target_data as xml) as TargetData

from sys.dm_xe_session_targets st

join sys.dm_xe_sessions s on s.address = st.event_session_address

where name = 'system_health') AS Data

CROSS APPLY TargetData.nodes ('RingBufferTarget/event[@name="xml_deadlock_report"]') AS XEventData (XEvent)

If that doesn't quite get it, try pulling the target_data into a xml variable and then working off of that:

declare @xml xml

select @xml = target_data

from sys.dm_xe_session_targets

join sys.dm_xe_sessions on event_session_address = address

where name = 'system_health'

select CAST(

REPLACE(

REPLACE(XEventData.XEvent.value('(data/value)[1]', 'varchar(max)'),

'<victim-list>', '<deadlock><victim-list>'),

'<process-list>','</victim-list><process-list>')

as xml) as DeadlockGraph

FROM

(select @xml as TargetData) AS Data

CROSS APPLY TargetData.nodes ('RingBufferTarget/event[@name="xml_deadlock_report"]') AS XEventData (XEvent)

Maxer SSCrazy Eights Points: 8964 More actions · Answer 13

Thank you, that second set of queries executed almost immediately!

It looks like they return zero rows though, so if I understand that correctly it means there have been no deadlocks since our last server restart?

Jonathan Kehayias One Orange Chip Points: 26778 More actions · Answer 14

Not necessarily, it means that there are none in the current data in the ring_buffer target. The ring_buffer is a in memory target and works in a FIFO manner once the 4MB of buffer space is full. Depending on how much data is in there it could mean that none have occured, I can't really tell you without seeing it.