Hello all, There is a new feature I would like a review on! https://github.com/apache/incubator-gossip/compare/master...edwardcapriolo:GOSSIP-42
I also have a general question for someone that knows math. As you can see here I had to introduce some fuzz here: https://github.com/apache/incubator-gossip/compare/master...edwardcapriolo:GOSSIP-42#diff-d224dbebf192fd1917452742052621f5R73 Commons math has a couple edge cases with our implementation. Negative values and 0's pose problems for the PHI calculations. These happen in a couple places: during NTP clock shifts etc. One thing I did was rejected any sample which was negative, The second issue I pointed out above is that there are some normal situations where variance blows up. IE if the timestamps are 5, 10 , 15. The variance is 0 and the method throws exception. I hacked this by adding a .1 in here. I also switched to nano-seconds because that makes the change of equally spaced timestamps more rare! IF anyone wants to think hard around these edge cases and make suggestion let me know. Right now the effect is the node stays "down" for a few cycles because with few samples these events are more likely.
