SisyphusDB is a strongly consistent distributed key-value store built from scratch in Go, designed for memory-constrained write-heavy workloads.
Storage Engine
An LSM-Tree engine with memtables for in-memory writes, SSTables for sorted durable storage, and Bloom Filters for fast negative key lookups. Compaction runs in the background to bound read amplification.
Consensus
Implements the Raft consensus algorithm for fault-tolerant log replication across nodes — leader election, log replication, and membership changes are all handled. Unit and integration test suites validate consensus and replication logic; chaos testing confirms leader election and log consistency guarantees under node failures.
Observability
Prometheus metrics and Grafana dashboards track cluster health, leader state transitions, P99 latency, and write throughput in real time.
Production Deployment
Deployed on AWS EKS using Kubernetes StatefulSets. Achieved 10,000+ write RPS with <55ms P50 latency and <90ms P90 latency, validated through load testing.
Open Source
Led as a GSSoC/Apatre project with 10+ contributors and multiple merged PRs. Also contributed a deadlock fix to PingCAP/TiKV (PR #10248) during this work.
Stack
Go · gRPC · Raft · Kubernetes · AWS EKS · Prometheus · Grafana