At the risk of sounding flippant, if it’s neither strictly CP or AP, it’s neither consistent nor available– it has the capacity to go down and mangle your data regardless of the steps you take to mitigate it. That’s a terrible recipe for data storage– bordering on useless unless you use it only as a secondary view into some data that’s store more robustly elsewhere in accordance with your storage and access needs.
‘Strictly’ is just a fluff word.
From my understanding of the post, they actually have 2 datastores: the AP one is where your user data goes in, the CP one is where metadata goes in. It seems like that store is only needed for topology changes. My guess is that there will be some lease-time on data so even partitioned nodes will keep working for awhile. I’m guessing there will be some funky failure scenarios, though.
The AP portion seems a lot like Cassandra, and the differences I’m not sure if the differences matter.
What’s the big win of using InfluxDB and their own clustering solution to just using an existing technology Cassandra?
I don’t have any experience running Cassandra but InfluxDB’s query API and ease of deployment are advantages IMO.