We’ve implemented Service Broker for handling a small portion of the transactions we want to catch and apply into our Operational Data Store. Mostly we wanted to make sure we accounted for actual DELETE operations and handled those records correctly. We can use Change Data Capture at this time on our source systems and Service Broker seemed to fit the bill. We still handle the remaining inserts and updates through SSIS.
While running Service Broker, we noticed a pretty constant CPU hum on the receiving server. That seemed odd, but we had a lot of trouble tracking it down. Regular Profiler traces didn’t show any running TSQL, “normal” Service Broker traces weren’t showing much, either. We just saw SQL Server running at a pretty constant 20% even with nothing seemingly happening.
I want to give a public thank you to Mark Hill (twitter) for doing a little extra digging and catching the root cause of this. In our tinkering with Service Broker, Queues, Activation, Stored Procs, and such we missed some very important information along the way.
We had written a stored procedure that would be used on the Receiving side of our Service Broker queue to run as the “Activated” proc. Inside the stored procedure, we included a WHILE loop to process anything that was in the queue. If nothing, exit. That seemed pretty simple – if something comes into the queue, process it and stop when nothing is left. End of story, right?
Sadly, this is where we missed a small, but important, fact about Service Broker. If you have an activated stored procedure that doesn’t use the WAITFOR command, Service Broker, we execute that stored proc as many times as possible looking for something to process. We looked at it, turned off the Activation on the queue and saw the CPU drop to nothing. We re-activated the queue and saw the CPU shoot up again. We tweaked the stored procedure after that and added a WAITFOR command with a timeout of 60000. After doing this, we saw the stored procedure run to process everything in the queue, then go idle, which was exactly the intended behavior in the first place.
I’m not going to pretend to be an expert on Service Broker. We had tried to code with the intention of being able to use this stored procedure as an Activated stored procedure or as one called from an external process to work through the queue. While that may be possible, it was unnecessary in our actual usage. Adding a WAITFOR command around our queue processing eliminated our extra, non-essential CPU usage and stopped trying to execute a stored procedure for no reason.
Resources
Pro SQL Server 2008 Service Broker ( Amazon | Apress )
WAITFOR (Also see this performance article – it’s the little details that get you)
