Please start any new threads on our new
site at https://forums.sqlteam.com. We've got lots of great SQL Server
experts to answer whatever question you can come up with.
| Author |
Topic |
|
Wanderer
Master Smack Fu Yak Hacker
1168 Posts |
Posted - 2008-07-09 : 05:43:44
|
| Anyone ever seen an "exchange event" as part of a deadlock graph?Running 64 bit SQL 2005 Ent, SP2, CU2.A stored procedure created over 2 years ago, and generally running successfully, against a datbase that support 24x7 web based trading, is (today) deadlocking.The SP running a ONLY a select (albeit a 10 table join - it's a data extract for central financial reporting, and thus must be accurate). No use of locking hints or deadlock priorities (although I ma make this a higher deadlock priority to get it through).The SP is typically called from a remote server, using a DTS package (old code - needs to migrate, I know), but when the SP is run local to the data, from SSMS, it is still a victim of a deadlock, with the same characteristics in the deadlock graphs seen in profiler.In the deadlock graph, I see that the SP seems to requesting, and owning "exchange event" - and so far google, BOL and SQLTeam seems to have nothing regarding "exchange event".We do not have any MS Exchange integration with the DB, and there are no triggers, CLR or anything else I can think off that might be obscure...Any ideas what this might be?*##* *##* *##* *##* Chaos, Disorder and Panic ... my work is done here! |
|
|
spirit1
Cybernetic Yak Master
11752 Posts |
Posted - 2008-07-09 : 07:22:59
|
| can you show us the graph?is this "exchange" in the EventSubClass column?_______________________________________________Causing trouble since 1980Blog: http://weblogs.sqlteam.com/mladenpSpeed up SSMS development: www.ssmstoolspack.com <- version 1.0 out! |
 |
|
|
Wanderer
Master Smack Fu Yak Hacker
1168 Posts |
Posted - 2008-07-09 : 08:07:03
|
| Not sure if I can show the graph - I guess I can screenshot it, but it is quite big (26 elipses representing processes, even though 25 of them are from my spid, and 1 from the other spid involved in the deadlock).Further to this, I noticed that this MIGHT be related to parallelism - I created a version of the SP with OPTION (MAXDOP 1), and that seems to run without deadlocks. Unfortunately, it does causes blocking on core tables, and the length of time it ran (over 3 minutes) started having on impact on certain customer traffic, so I had to stop it.I'm looking at the external systems to determine whether we can run this during off-peak hours (we usually run at 03:00, but that run failed this morning). Then I plan to check the CU3-8 fix lists to look for reference to parallelism/other fixes that might be related - we already have CU7 on a few servers, and might go that route...*##* *##* *##* *##* Chaos, Disorder and Panic ... my work is done here! |
 |
|
|
|
|
|
|
|