FeatherCast

Cassandra has been deployed for many years at Instagram and is still growing fast. Over the years, we’ve constantly improved the design of our infrastructure: we’ve introduced Cassandra proxy nodes to decouple the processing and storage workloads, we’ve developed Rocksandra, a new storage engine relying on RocksDB that reduced the GC pressure and improved the efficiency of our clusters. Lately, we’ve added an intermediate layer in our Cassandra infrastructure, where we could add many traffic optimizations to further improve our Cassandra clusters’ efficiency and reliability. This talk will explain in detail this new component and the results we’ve observed.