Here’s something to waste your time and bore your brains out.

There are three nodes, say Nodes A, B and C. Node A and C are application servers. Node B is a messenger for nodes A and C. Although Node B is not needed but for reasons I’m not allowed to discuss and would rather not discuss, it has to be there and is the root cause why we have a problem.

Let’s start with an example series of events. A new transaction begins from Node C which sends data to Node B. Node B stores the data in a queue. Node A fetches x items from the queue every y seconds.

After Node A processes the transaction, it begins a new transaction. Node A sends data to Node B only this time, it is no longer stored in a queue. Node B immediately forwards the data to Node C.

On some occasions, Node C starts a new transaction as an acknowledgment from the data that originated from Node A. This transaction is treated as a transaction similar to the first one only that Node A knows that it’s an acknowledgment.

Originally, Node A and Node C communicated to each other directly. The queue didn’t exist. The queue came in to fix a couple of problems but it introduced new ones. Recently, the script in Node A that polls data from Node B died and now it’s catching up with backlog. The script processed data one by one. This showed how slow the main application in Node A is.

I already have a solution in mind and I enjoy writing boring stuff but I’d like to hear other views.

This entry was posted on Thursday, April 24th, 2008 at 21.36.36 and is filed under Software/Code, Techitude, foo, bar and h4xx1ng. You can follow any responses to this entry through the RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.