This tool follows a twitter stream and prints out the most frequently encountered words over the given reporting frequency.

This is related to a concept we gave to a customer who wanted to solve this problem a while back. This is a significantly scaled down version, but it can exercise a server pretty hard by doing the following:

Every tweet that comes in is split into the words that make it up and those words are counted within the tweet. Each word is then incremented against the current time window.

Also after each tweet, we maintain the top 1,000 words. In order to do this, we do a cas loop where we get the current list, do a giant multi-get of the union of all of the words currently in the list and all of the words we just found to find their current counts, then update the list with this value. The cas frequently fails, and will likely fail more with more -workers.

Based on the -interval parameter, a report will be sent to stdout showing the top of the top list. That's currently just a simple get and in-memory ops.

In general, twitter only provides a 1% stream of tweets. Using the -multiplier parameter, you can send every tweet through multiple times such that -multipler=100 should approximately equal the traffic of all of twitter (though the words being processed would vary, etc...).