wrk2

wrk2
Name	wrk2
Title	wrk2
Developer	John Resig; Glen Keane; community contributors
Released	2013
Programming language	C (programming language); Lua (programming language)
Operating system	Linux; macOS; FreeBSD
Genre	load testing tool
License	MIT License; open-source software

wrk2

wrk2 is a constant-throughput HTTP benchmarking tool derived from the wrk project that targets stable request rates for latency-focused load testing. It is used by practitioners in web performance, site reliability engineering, and systems research to evaluate latency, throughput, and queueing behavior across stacks such as NGINX, Apache HTTP Server, Envoy, and HAProxy. The tool is commonly applied in studies alongside platforms like Kubernetes, Docker, and cloud providers including Amazon Web Services, Google Cloud Platform, and Microsoft Azure.

Overview

wrk2 provides a mechanism to generate HTTP load with a controlled, constant request rate to better measure tail latency and system behavior under steady-state load. It complements other benchmarking tools such as ApacheBench, JMeter, Gatling, Siege, and Locust by focusing on rate control rather than bursty throughput. Researchers and engineers use wrk2 to produce reproducible results for performance studies involving stacks like Redis, PostgreSQL, MySQL, and proxy layers such as Varnish.

Features and Design

wrk2's principal feature is a closed-loop, scheduled request generator that enforces a target requests-per-second rate using user-space timers and pacing algorithms influenced by research from institutions like Stanford University, MIT, and UC Berkeley. It supports scripting of requests via Lua (programming language), enabling complex scenarios that interact with systems including Consul (software), etcd, Prometheus, and Grafana. The codebase leverages high-performance primitives from pthreads, epoll, and kqueue to scale across CPU cores, and interoperates with TLS stacks such as OpenSSL, LibreSSL, and BoringSSL for secure endpoints like Let’s Encrypt-protected services.

Usage and Examples

Typical usage patterns involve configuring target endpoints (for example services like NGINX frontends), specifying constant rates, and providing Lua request scripts that model realistic client behavior against application servers such as Node.js, Django, Ruby on Rails, Spring Framework, and ASP.NET Core. Example workflows include pairing wrk2 with observability tools like Jaeger (software), Zipkin, and New Relic to correlate request latencies with resource metrics from Prometheus exporters or Datadog. Operators frequently run experiments in controlled environments such as Vagrant boxes, VirtualBox, or cloud VMs to evaluate autoscaling policies for platforms like Kubernetes Horizontal Pod Autoscaler or AWS Auto Scaling groups.

Performance and Benchmarking Methodology

wrk2 encourages experimental rigor by enabling fixed-rate load inputs that reduce confounding variability when measuring tail percentiles (p50, p95, p99) and queuing effects in middleware like Envoy (software) or HAProxy. Benchmarks typically reference best practices promoted by organizations such as USENIX, ACM SIGCOMM, and IEEE conferences, and are compared against baseline tools such as wrk (software), hey, and bombardier. Researchers analyze results using statistical packages from R (programming language), Python (programming language), and libraries like NumPy, pandas, and visualization via Matplotlib or Grafana dashboards.

Implementation and Architecture

wrk2's architecture builds on a multi-threaded event loop with user-space pacing to maintain constant request rates, integrating a Lua interpreter for request generation logic similar to approaches used in HAProxy and Nginx module ecosystems. The implementation relies on low-level networking APIs available on Linux, macOS, and FreeBSD, and interacts with TLS libraries including OpenSSL for HTTPS workloads. The project structure and patch contributions often follow workflows common to projects hosted on platforms like GitHub, with continuous integration using services such as Travis CI and GitHub Actions.

Community and Development History

The project emerged as a fork and enhancement of existing load generators to meet the needs of practitioners focused on latency tail behavior, drawing contributors from companies and institutions like Facebook, Google, Netflix, Twitter, Cloudflare, and academic labs. Development discussions and issue tracking typically occur on GitHub issues and mailing lists influenced by open-source collaboration patterns seen in projects like Linux kernel and OpenStack. The community maintains integrations, benchmarking scripts, and example repositories that demonstrate wrk2 usage with ecosystems including Kubernetes, Docker Compose, and service meshes like Istio.

Category:Software