The Power of Lacinia & Hystrix in Production

This repo has the code used for the talk The Power of Lacinia & Hystrix in Production initially prepared for the Clojure/conj 2017.

Table of Contents

• Organization
• Presentation
• Running The Demo Project
• Simulating Latency
• Load Testing

Organization

• e-commerce-demo - contains the demo project that was used for this presentation
• statistics - contains the scripts for running the load tests as well as a dump of the test results used on the talk

Presentation

Here's the recorded video from Clojure/conj 2017:

The deck can be found here.

Running The Demo Project

Refer to the README on e-commerce-demo

Simulating Latency

The e-commerce-demo has a simple embedded latency simulator that can be configured at e-commerce-demo.application.

The application gets initialized with two simulators and their settings are very straight-forward:

   :delayer-prods-config {:mean 100
                          :std-dev 100}

Load Testing

I've used Gil Tene's great wrk2 as a HTTP benchmarking tool. It produces a constant throughput load and accuratly calculate latency details to the high 9s.

To build it for MacOS make sure to have XCode Command Line Tools and brew installed. Then you'll need to have openssl and expat installed:

$ brew install openssl
$ brew install expat

Mac's make needs a bit help finding the libs:

export LDFLAGS="-L/usr/local/opt/openssl/lib -L/usr/local/lib -L/usr/local/opt/expat/lib"
export CFLAGS="-I/usr/local/opt/openssl/include/ -I/usr/local/include -I/usr/local/opt/expat/include"
export CPPFLAGS="-I/usr/local/opt/openssl/include/ -I/usr/local/include -I/usr/local/opt/expat/include"
export CPATH=/usr/local/opt/openssl/include:"${CPATH}"

make

Then you can run the load script against your localhost with:

$ cd statistics/
$ wrk -t8 -c128 -d2m -R100 -s scripts/post.lua http://localhost:3000/graphql\?

The latency report will be printed after 2 minutes (-d parameter above).

Notice that wrk2 runs its sampling initializer for 10 seconds and therefore shorter runs tend to be more imprecise.

The constant rate parameter (-R) may also have different implications based on your system's capabilities.

License

Copyright © 2017 Tiago Luchini

Distributed under the MIT License.