Posted by David Boike on April 30, 2011
Sometimes I need a rolling counter, especially in diagnostics-related scenarios. How many requests have occurred in the last hour? It’s not okay for a counter object to drop out of cache every hour, because the value will be meaningless if I happen to observe it at (Cache Drop + 3 minutes).
A real rolling counter is needed in these situations. The counter must increment, and then at some point those hits must drop off.
But especially in these situations, low impact is the key. A Queue where each item contains a timestamp is too unruly. Too many objects are created and too much cleanup is required. Less is more.
Read more »
Posted by David Boike on April 12, 2011
When presenting my Distributed Application Development with NServiceBus talk at the Twin Cities Code Camp 10, I was asked by an attendee why NServiceBus’s automated retry feature doesn’t have some sort of a delay or back-off algorithm. Unfortunately I had never really thought about it, and I didn’t have a very good answer for him.
This week I have the good fortune to be attending Udi Dahan’s Advanced Distributed System Design course in New York, so I thought I would get the answer direct from the source.
Udi gave me several good reasons why this is intentionally left out of NServiceBus. I’ll try to convey his answers and add my own thoughts as well.
Read more »