Skip to main content

ScyllaDB Solves Cassandra Timeouts and Shrinks Clusters at Fanatics

Niraj Kothari discusses many benefits of migrating order capture from Cassandra to ScyllaDB. [Read the complete case study]


Hello everyone. My name is Niraj Kothari. I am the lead for Platforms Engineering at Fanatics. Fanatics is an ecommerce company in the niche market for sports merchandising.

The problem that we were trying to solve was we were using Cassandra as our main order capture database and due to a certain type of data, like cart mutations, it was giving us a lot of issues related to timeouts. The reason for the timeouts was huge data sets sometimes caused pressure on the hip due to GC pauses as well as CPU spikes that caused timeouts for our clients.

Moving to ScyllaDB really helped us because it solved all these issues. There are no GC pauses and the latencies were pretty decent in ScyllaDB. And in the very end, it actually helped us reduce our stack size from 55 nodes of Cassandra nodes to only 12 nodes of Cassandra with three nodes of ScyllaDB. This not only helped us solve our problem, it actually reduced our TCO which is total cost of ownership at Fanatics. 

When you have a big cluster, like 50 nodes, maintaining the cluster, doing a rolling restart on the big cluster takes a long time. And this actually helped us to reduce that as well. With just a 12 node cluster, we can actually do a rolling restart within less than an hour. Before it used to take five to eight hours. So this not only reduced TCO, (total cost of ownership) but it also reduced the pain that the database engineering team was taking to actually maintain the cluster in a healthy state.

Overall, ScyllaDB has been great. Latencies are awesome. Read and Write latencies are below two milliseconds and we are so far pretty happy with ScyllaDB.

If someone is getting into ScyllaDB for the first time, I highly recommend taking the enterprise version. We got a lot of support around it and so far we have been really happy with the kind of support we got from ScyllaDB support. It has been great.