tp-load-balancer

Load balancer service is a REST service which expose one endpoint

• /route?id=${userid} - takes userid and return one of the weighted group defined in externally delivered properties file
  

Building

For generate jar file run

• mvn package

Running application

Application is spring boot repacked jar, which can be run using command:

• java -jar TP-LOAD-BALANCER-1.0-SNAPSHOT.jar --spring.config.location=file:{externally.delivered.properties}

Properties format

Weighted groups should be defined as a touple in format

• groupsWeights={GroupName:'number', .... } - where number is percentage distribution of requests among groups divided by ten
  Properties should also specify consistent hashing buckets number, which currently have to has value 10.
• numberOfBuckets=10
  Example of properties file content:
  groupsWeights={GroupA:'2',GroupB:'2',GroupC:'4',GroupD:'2'}
  numberOfBuckets=10

Dockerization

Project has Dockerfile delivered. following command creates docker image:

• docker build -t docker_image_tag .
  

Then run docker image with command:

• docker run -d -p 8080:8080 docker_image_tag
  

When running on Windows service should be reached using address retrieved by command:

• docker-machine ip default

Usage

While service is running it can be reached with wget

• wget http://${ip}:8080/route?id={alphanumeric_id}

Testing

Project contains functional main.test which verifies requests distribution among predefined, weighted groups
FunctionalTestRunner - periodicaly reports requests distribution among groups.
When running main.test it should also have --spring.config.location=file:{externally.delivered.properties} passed as a program parameter. Example of report:
functional.config.FakeClient : 20.346608898516912 percentage of traffic goes to group GroupA
functional.config.FakeClient : 20.113314447592067 percentage of traffic goes to group GroupB
functional.config.FakeClient : 39.27678720213298 percentage of traffic goes to group GroupC
functional.config.FakeClient : 20.32994500916514 percentage of traffic goes to group GroupD

Performance tests

For performance tests wrk application was used. It can be retrieved from https://github.com/wg/wrk.
In order to achieve different user id for each requests loadtest.lua script is delivered.

Example of usage:

• wrk -t 2 -c 2 -d test_duration -s /path/to/loadtest.lua http://${ip}:8080
  

Performance tests results

Five main.test runs was performed on:
CPU: Intel Core i5 72000
2 Cores 2,5GHz
RAM: 8GB

results: https://docs.google.com/spreadsheets/d/1XC8VOSM4L2ZlT0494aP55feKyJ5BPECpg0vcNsmh1d4/edit?usp=sharing

• Average throughput was: 11909 Req/sec
• Average latency was 521.244 micro seconds