Pushnami

At Pushnami we were using Redis for quite a long time, and still do for disposable caches. However as we continued to grow, we started to feel limitations of keeping data sets in memory, and only being able to use 1 CPU per server. We started to look around. Our requirements were pretty clear: - Very fast reads and writes (sub ms) - Ability to hold Terabytes of data - Ability to horizontally scale - Strong support for TTL and data eviction After looking at several alternatives we decided on Scylla , which had familiar Cassandra interface but claimed to be much faster. 2 years later - we run a cluster of 6 nodes that holds over 20 TB of data, and performs 150,000 operations per second. I would say pretty impressive! It was also easy to get started on AWS, as ScyllaDB provides prebuilt AMI with optimal configurations baked in.

At Pushnami we were looking at several alternative databases that would support following architectural requirements: - very quick prototyping for an unknown domain - ability to support large amounts of data - native ability to replicate and fail over - full stack approach for Node.js development After careful consideration MongoDB came on top, and 3 years later we are still very happy with that decision. Currently we keep almost 2TB of data in our cluster, and start thinking about sharding.