electric-sql/ds-bench

This project is a benchmarking tool built to test and compare a specific kind of server called a durable stream server. A durable stream is a system that reliably stores a continuous flow of data, like a log of events, so that clients can write new entries and reliably read or replay them later, even after a crash. This tool measures how fast and how stable different server implementations of that idea are under load. Instead of testing just one server, it is designed to run the same workload fairly against several different systems, including one called durable-streams written in Rust, a Node.js reference server, a project called ursula, and one called S2. Each workload is described in a JSON configuration file that defines what the test does, such as measuring how many writes per second a server can sustain before it hits its limit, how memory and response times behave when load is held steady for a long time, or how quickly many clients can catch up on a stream after reconnecting. The tool runs on Kubernetes, either on a small local cluster using a tool called kind for quick testing, or on Google Cloud's GKE for larger, more realistic runs. It automatically sets up the cluster, deploys the server being tested, runs a fleet of client machines to generate load, collects detailed timing statistics from every client, and merges them into one clear picture of performance. When a full test run finishes, the cluster tears itself down automatically to avoid unnecessary cloud costs. This is a specialized tool for engineers who build or evaluate infrastructure software, particularly anyone deciding which streaming or event-log server to use in a production system. It is not something a typical app developer would use day to day, but it produces reports that make it easier to compare real-world tradeoffs between different backend technologies.