Chasing the long tail of Clojure HTTP servers.
This work is not intended as a critique to the excellent work that the library authors have done. Each of the HTTP server libraries compared here have excellent performance characteristics, they are very stable and widely used to power loads of HTTP based services all around the world.
This project aims to examine the long tail for a very specific case. The case we are going to look here fits the model of a service-to-service communication for a backend service. Where we assume long running persistent connections, and continuous throughput of requests. This particular case is very different from front facing services with bursty and erratic request throughput.
Therefore before jumping into conclusions, please think about the specifics of your case and if unsure collect metering data and use this or other test harnesses on your services to verify the behaviour of these libraries in your particular case.
The aim of the long tail analysis is to understand the impact of the library in a vanilla case. Settings and configuration might not apply in your project when extensive business logic is applied which substantially change the scenario.
NOTE: I'm aware of the similar work from Peter Taoussanis clojure-web-server-benchmarks While Peter's work is targeted to find the peak performances in a number of different setup, my work only looks at the tail latency for a specific case.
- See latest report
- All reports
To run the tests yourself you need a few packages to be installed and available in the PATH:
- Install the Java JDK
- Install leingen
- Install wrk2
- Install neofetch
- Install planck-repl
then clone the repo:
git clone ...
cd clojure..
./bin/run-all.plk
by default it runs a the test for 5 minutes with a warm up time of 10
seconds. To change these values create a file called config.edn
with the following content:
{:duration "3h" :warm-up-time "120s"}
then run it with:
./bin/run-all.plk -c config.edn
The script will build and run all the tests for all the servers.
Once completed the resulting data will be available in the folder ./results
under a folder with the timestamp of when the test was started.
You can generate the graphs with:
pip3 install --user --upgrade hdr-plot
./bin/plot-all.sh ./results/yyyy-mm-dd_hh-mm-ss/
Copyright © 2018 Bruno Bonacci - Distributed under the Apache License v 2.0