First of all thanks for providing a great product. Currently we are evaluating the server for our own metrics and ran into a snag. Any dataset with more than 32,767 matched rows will result in a out of bounds error. We went ahead and changed row_index from a short to an int and also change some of the other variables that we got loss or precision errors into an int. This fixed the issue for us and we can run results for all 14 years of data. I'd submit a patch but I am not well versed in Java and don't want to miss any glaring details. Thanks again!
Go ahead and issue a pull request so you can get credit for the fix. I can review and help clean it up. And please sign http://opentsdb.net/contributing.html if you haven't already.