Why Every Automation Tester Should Learn Kubernetes in 2026

# e2e-job.yaml - runs a pytest + Selenium suite once
apiVersion: batch/v1
kind: Job
metadata:
  name: e2e-tests
spec:
  backoffLimit: 0
  activeDeadlineSeconds: 1800
  ttlSecondsAfterFinished: 7200
  template:
    spec:
      restartPolicy: Never
      containers:
        - name: tests
          image: registry.example.com/qa/e2e-tests:1.3.0
          command: ["pytest", "tests/e2e", "-q", "--junitxml=/reports/results.xml"]
          env:
            - name: BASE_URL
              value: http://shop-frontend.qa.svc.cluster.local
            - name: SELENIUM_REMOTE_URL
              value: http://selenium-hub.qa.svc.cluster.local:4444
          resources:
            requests: { cpu: 500m, memory: 512Mi }

# Start the run
kubectl apply -f e2e-job.yaml -n qa

# Watch the pod being created and finishing
kubectl get pods -n qa -l job-name=e2e-tests -w

# Follow the test output
kubectl logs -f job/e2e-tests -n qa

# Wait for success (use in CI; returns non-zero on timeout)
kubectl wait --for=condition=complete job/e2e-tests \
  -n qa --timeout=30m

# Did it pass?
kubectl get job e2e-tests -n qa \
  -o jsonpath='{.status.succeeded}{" "}{.status.failed}'

# playwright-shards.yaml
apiVersion: batch/v1
kind: Job
metadata:
  name: pw-regression
spec:
  completionMode: Indexed
  completions: 4
  parallelism: 4
  backoffLimit: 0
  template:
    spec:
      restartPolicy: Never
      containers:
        - name: playwright
          # Official image already contains the browsers
          image: registry.example.com/qa/pw-tests:2.1.0
          command: ["sh", "-c"]
          args:
            - >
              npx playwright test
              --shard=$((JOB_COMPLETION_INDEX + 1))/4
              --reporter=blob
          env:
            - name: BASE_URL
              value: http://shop-frontend.qa.svc.cluster.local
          resources:
            requests: { cpu: "1", memory: 1Gi }

# Dockerfile for the test image (based on the official Playwright image)
# FROM mcr.microsoft.com/playwright:v1.48.0-noble
# WORKDIR /tests
# COPY package*.json ./
# RUN npm ci
# COPY . .

# See each shard's output
kubectl logs -n qa -l job-name=pw-regression --prefix

# Blob reports from every shard must be copied to one place
# (shared volume or object storage), then merged:
npx playwright merge-reports --reporter=html ./all-blob-reports

# Check shard status: one pod per index
kubectl get pods -n qa -l job-name=pw-regression \
  -o custom-columns=POD:.metadata.name,STATUS:.status.phase

# Install Selenium Grid with the official Helm chart
helm repo add docker-selenium https://www.selenium.dev/docker-selenium
helm repo update
helm install selenium-grid docker-selenium/selenium-grid -n qa

# Find the Grid service name and port
kubectl get svc -n qa

# Python test pointing at the Grid inside the cluster
from selenium import webdriver

opts = webdriver.ChromeOptions()
driver = webdriver.Remote(
    command_executor="http://selenium-hub.qa.svc.cluster.local:4444",
    options=opts,
)
driver.get("http://shop-frontend.qa.svc.cluster.local")
assert "Shop" in driver.title
driver.quit()

#!/usr/bin/env bash
# ephemeral-env.sh <pr-number> <image-tag>
set -euo pipefail
NS="pr-$1"
TAG="$2"

kubectl create namespace "$NS"
# Always clean up, even if tests fail
trap 'kubectl delete namespace "$NS" --wait=false' EXIT

helm install shop ./charts/shop -n "$NS" --set image.tag="$TAG"
kubectl rollout status deployment/shop-frontend -n "$NS" --timeout=5m

# Point the app URL at this namespace; keep the shared Grid in qa
sed "s/shop-frontend.qa.svc/shop-frontend.$NS.svc/" e2e-job.yaml | kubectl apply -n "$NS" -f -
kubectl wait --for=condition=complete job/e2e-tests -n "$NS" --timeout=30m
kubectl logs job/e2e-tests -n "$NS" > "e2e-$NS.log"

SUITE READINESS FOR KUBERNETES

Packaging
[ ] Test code and dependencies in one container image
[ ] Image tag pinned (no 'latest')
[ ] Base URL and Grid URL read from environment variables
[ ] No hard-coded localhost addresses

Independence (needed for sharding)
[ ] Each test creates its own data or uses unique users
[ ] Suite passes in random order
[ ] No test depends on another test's output

Reporting
[ ] Exit code non-zero when any test fails
[ ] JUnit XML or blob reports written to a known path
[ ] Reports copied to storage the CI can read

Resources
[ ] CPU and memory requests set for each pod
[ ] activeDeadlineSeconds set to stop hung runs

# .github/workflows/e2e.yml (CI step that runs tests in the cluster)
name: e2e
on: pull_request

jobs:
  e2e:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Configure cluster access
        run: echo "${{ secrets.KUBECONFIG_QA }}" > kubeconfig
      - name: Run tests in a fresh namespace
        env:
          KUBECONFIG: kubeconfig
        run: ./ephemeral-env.sh ${{ github.event.number }} ${{ github.sha }}
      - name: Save logs
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: e2e-logs
          path: e2e-*.log

FLAKINESS LOG - track tests that behave differently in the cluster

Date | Test name | Shard | Failure | Root cause | Fix
-----|-----------|-------|---------|------------|----
     |           |       |         |            |

Common causes to check:
- Fixed sleeps instead of waiting for elements
- Shared login account used by two shards
- Pod CPU limit too low, browser renders slowly
- App pod not Ready when tests started
- Timezone or locale different in container

Rule: a test that fails twice in this log without a
product bug gets fixed or quarantined before the next run.