Definitely not null keys. They are time based UUIDs. Basically the test set I’m running is a collection of articles stored in Cassandra and their key is the uuid generated when inserted there.
Get the articles from bing api and its the same set that bing returns in both cases (same number (67) of articles and same articles). So my theory were that the time based UUIDs where so similar that the hash and modulo ended up being the same. But after reading your responses I’m back at just being puzzled. I guess another approach could be to generate a random uuid and use that for the message record key instead?