Wednesday, May 11, 2005

"Zombie" scenario in Aggregator

In my current project we used Aggregator to batch the files after splitting. We used combination of "Activate" subscription and "Correlation" Subscription. The second receive shape (for Correlation) has a listener that had delay of 60 seconds. We found that in some large files with more that 200 transactions there was some transaction loss. There was no trace on HAT or Event vwr. After thorough investigation we found out that the Orchestration was terminating without waiting for all the messgages to be be consumed from the Message Box. Basically, some messages were not being consumed and hence discarded. This is called "zombie" scenario. There is a nice article on this at:

Solution: It was simple. We increased the timer delay from 60 seconds to 240 seconds. It did effect the performance but that was offset by the "no loss" of transactions.

No comments: