K3s – Minimal Configuration and Observations on a Single VM

1- (2 vCPU / 2 GB RAM)

Test Scenarios

• Simple Deployment (Nginx)

  • A web server Nginx deployed through K3s runs correctly.
  • Exposed using NodePort: accessible and stable, even under continuous load testing on the index page.

• Connection to Rancher

  • The K3s cluster was added to Rancher without any issues.
  • Initial monitoring and management were operational.

Problems Encountered

When deploying the Boutique (microservices-demo) application:

• Degraded Performance

  • Very slow deployment (≈20 min instead of <1 min on a 3-node K3s cluster).
  • Low responsiveness: kubectl console and Rancher interface are sluggish.

• Unstable Services

  • Some pods remain in the NotReady state for a long time.
  • Random and automatic restarts of services (CrashLoopBackOff observed).

• Critical Memory Usage

  free -m
                total  used  free  shared  buff/cache  available
  Mem:          1966    1659    85     18       424         307
  Swap:            0       0     0
  

2- VM 3 vCPU / 2.5 GB RAM

Test Scenarios

• Deployment of an Nginx Server

  • One Nginx pod deployed via K3s works properly.
  • Exposed using NodePort → accessible and stable under repeated requests.

• Rancher Integration

  • The cluster was added to Rancher without any problems.
  • Viewing statistics works (some latency, but functional).

• Deployment of the Boutique (microservices-demo)

  • The application starts and runs.
  • Some slowness observed, but overall usable.
  • The MetalLB LoadBalancer worked well to expose the Boutique application.

Load Test (LoadGenerator)

Initial Configuration

kubectl -n default get deploy loadgenerator -o yaml | yq '.spec.template.spec.containers[0].env'

FRONTEND_ADDR: frontend:80
USERS: "10"
RATE: "1"

I increase the load:

# 1: 50 users, 5/s --> OK
kubectl set env deploy/loadgenerator USERS=50 RATE=5 

# 2: 200 users, 20/s --> Slower and less responsive
kubectl set env deploy/loadgenerator USERS=200 RATE=20

# 3: 500 users, 50/s --> Much slower, services stop responding, console becomes slow
kubectl set env deploy/loadgenerator USERS=500 RATE=50

3- VM 3 vCPU / 3 GB RAM

• Deployment of the Boutique (microservices-demo)
  • The application starts and runs.
  • Some slowness observed, but overall usable.

K3s – Minimal Configuration and Observations on 3 VMs

1- 3 VMs with 2+0.5x2 = 3 vCPU / 2 + 2*512 = 3 GB RAM

Test Scenarios

• Deployment of the Boutique (microservices-demo)

  • The application starts and runs.
  • Some slowness observed, but overall usable.
  • The MetalLB LoadBalancer worked well to expose the Boutique application.

• Rancher

  • The cluster was added to Rancher, and latency increased.
  • Increasing load:

    # 1: 50 users, 5/s 
    kubectl set env deploy/loadgenerator USERS=50 RATE=5 
    

  • Rancher occasionally reports broken nodes!

2- 3 VMs with 2+1x2 = 4 vCPU / 2 + 2x1 = 4 GB RAM

Test Scenarios

• Deployment of the Boutique (microservices-demo)

  • The application starts and runs.
  • The MetalLB LoadBalancer worked well to expose the Boutique application.

• Rancher

  • The cluster was added to Rancher
  • Increasing load:

    # 1: 50 users, 5/s --> OK
    kubectl set env deploy/loadgenerator USERS=50 RATE=5 
    # 2: 200 users, 20/s --> OK with some latency 
    kubectl set env deploy/loadgenerator USERS=200 RATE=20
    # 3: 500 users, 50/s --> OK but noticeably slower 
    kubectl set env deploy/loadgenerator USERS=500 RATE=50