Gorilla: A fast, scalable, in-memory time series database
Summary
Gorilla is a fast, scalable in-memory time series database used for monitoring at Facebook. It stores billions of time-series in memory with a write-through architecture on top of HBase, uses a novel two-hour data block compression for timestamps and a XOR-based scheme for values to achieve large in-memory retention, and demonstrates horizontal scalability through sharding and replication. The article also discusses design goals, performance targets, and the practical outcomes of fault tolerance and read latency.