Introduction

Last week thanks to this post of Jean-Paul Smit I discovered Locust, a simple and effective tool to perform load tests.

You can already find different examples of IaC scripts to deploy an AzureContainerInstances-based infrastructure to run Locust. Here some examples.

Scenario and Technologies	Link
Headless Locust deployed with az CLI + ARM template	GitHub
Headless Locust deployed with az CLI + ARM with VNET connectivity (via appgw)	GitHub
Web UI Locust deployed with Terraform	GitHub
Headless Locust deployed with Terraform	GitHub

The goal of this exercise is to deploy a Locust WebUI test infrastructure using Bicep.

• a secondary goal was to not use additional scripts (bicep only) but unfortunately ARM do not provide the functionality to copy a file (the python script) to the file share. The ARM DeploymentStripts is used for that purpose.
• DeploymentStripts introduces a lot of un-necessary complexity. It's crazy to have to create MSI, ACI and a storage to simply upload a file to a share (the same task is extremely simple in Terraform or using the az cli).

Deployment

• create a resource group.
• run the az deployment command to target the bicep orcestrator file (00main.bicep) specifying the following parameters.
  • appId : prefix to name the different artifacts.
  • targetHost : hostname of the loadtest target
  • nrOfWorkers : number of locust workers [default value: 4]
  • environment [default value : dev]
  • randomizeLocation : whether to randomize the locations of the locust workers [default value : true

The locust master and all the support artifacts (storage account, deployment scripts, etc..) are created in the same location of the resource group.

$rgname = 'rg-laodtest'
az group create -n $rgname -l "francecentral"

az deployment group create -f ./00main.bicep -g $rgname 
    -p appId=lts 
    -p targetHost=lnl-webfront-afd.azurefd.net

result of the deployment

az cli : stop the workers

$rgname = 'ah-loadtest-with-aci'
$workers_names=(az container list --resource-group $rgname --query "[?contains(@.name, 'worker')].name" --output tsv)
foreach ($aci in $workers_names) {
    Write-Output "stopping the instance $aci"
    az container stop --name $aci --resource-group $rgname
}

Run Load Test

The Locus web interface is available at this URL : http://{appId }-{environment}-locust-master.{location}.azurecontainer.io:8089/ (http://lts-dev-locust-master.francecentral.azurecontainer.io:8089/)

For this exercise, I used Locust to test a AFD+APIM+APPSVC infrastructure.

• Azure Front Door (Classic - Global service, WAF enabled)
• Azure API Management (Premium - France Central - 100% logging to Application Insights)
• Azure App Service (3x P1V2 - France Central)
• Locust (8 workers : 'francecentral', 'japaneast', 'eastus2', 'westeurope', 'northeurope', 'westus', 'australiaeast', 'southcentralus')

3 run has been triggered with 2000, 400 and 800 users.

Northeurope have the better total latency (avg)

The metrics are split by 'Client Country'. 3 workers are in the US.

I was not expecting this latency from AFD

Avg duration per component during the run with 800 users.

APIM cpu (10%) during the run with 800 users

Conclusions

Locust on ACI enables you to create a distributed load test architecture in minutes. I was expecting a lower latency from FrontDoor, the test has been triggered with the WAF enabled using a single GET method. This article should be updated with a APIM+AppSvc baseline.