Time series analysis. The built-in vector operations are extremely fast. Also with the q language you can code up any customized analytical ideas quickly.
The database are all file based, very easy to maintain.
Very solid and fast interface to websocket, so you can interface with javascript easily.
Highly available: If nodes go offline for any reason, the system still operates.
Highly scalable: There is a minimum of 5 nodes, which can handle a lot by themselves. When scaling is required, it can be done easily, with minimal to no downtime on large scales.
Very fast searching: Riak has SOLR indexing built-into the core product, which makes querying for data very fast.
Deletes!!! We've seen on numerous occasions where Riak has "resurrected" deleted data. We've worked with Basho numerous times and tried multiple changes to the way we interact with Riak to prevent the problem but it still remains. The deletes seem to reappear weeks, even months, after the delete was issued. We've had to work around this issue by providing a "deleted" flag for all data objects stored in Riak. Thus, we do no delete but simply flip the flag. Excess baggage we would really like to not have to worry about.
Search. Currently there's no way to tell what data you have in Riak without already knowing a particular bucket/key. There is a way to list the keys for a given bucket but due to performance implications, this is not a viable method to lookup data. Especially when you have a large amount of keys in the bucket.
Right now, I'm on a project where we need databases that can run on embedded systems. Riak isn't necessarily the best fit for that scenario. But when we need a clustered database, that's where we'd start considering Riak.
Despite Basho going bankrupt and the project becoming fully open-source, community support is reasonably good, albeit a little slow at times. Paid enterprise-grade support is also available from former Basho engineers but the same company also contributes to the community support for free for basic questions or specific knowledge areas.
Python is very commonly used for large data analysis and in general is much easier to pickup than kdb+. The biggest drawback of kdb+ is the learning curve.
Because of the RESTful HTTP interface, the consistency model, and because of the catalog-driven data model, Riak was an easy win over Redis and Memcached.
Riak has been a key part of our company's build process for our client's search backend. It is valuable for is in that it provides a reliable way to view the current search index.
