Skip to content
Hands-on IT IT Drills

Appendix B: Final Exam

Goal: Prove that you can build and repair a Gateway with routes, TLS, policies and migration on your own · Duration: 120 min hands-on + 20 min questions · Prerequisite: Day 14

Rules

  • Cluster: kind gw-lab on the laptop, state after Day 7 (Envoy Gateway v1.9.2, cert-manager with ClusterIssuer lab-ca, payments-api from Day 7). cloud-provider-kind runs in the second terminal.
  • Everything attaches to a new Gateway infra/pruefung with its own IP. Hostnames: *.pruefung.gw.localtest.me. The lab objects from Days 1–7 stay untouched.
  • Allowed: kubectl explain, --help, Appendix A, gateway-api.sigs.k8s.io and gateway.envoyproxy.io. Not allowed: copying solutions from the chapters.
  • A task only counts if the check script reports it with ✔.
  • Passed: at least 70 of 100 points and Task 8 fully solved.

Fresh lab? Then in this order (commands from Days 0, 4 and 7):

~/gw-lab/code/demo/up.sh
helm upgrade --install cert-manager oci://quay.io/jetstack/charts/cert-manager \
  --version v1.21.2 -n cert-manager --create-namespace \
  --set crds.enabled=true --set config.gatewayAPI.enabled=true --wait
kubectl apply -f ~/gw-lab/code/tag04/lab-ca.yaml
kubectl apply -f ~/gw-lab/code/tag07/payments-api.yaml

Start:

mkdir -p ~/gw-lab/pruefung && cd ~/gw-lab/pruefung
export PATH="$HOME/gw-lab/code/demo:$PATH" CA=~/gw-lab/tag04/lab-ca.crt
kubectl config use-context kind-gw-lab
kubectl wait clusterissuer/lab-ca --for=condition=Ready --timeout=10s
kubectl -n payments get deploy payments-api

Expected:

Switched to context "kind-gw-lab".
clusterissuer.cert-manager.io/lab-ca condition met
NAME           READY   UP-TO-DATE   AVAILABLE   AGE
payments-api   1/1     1            1           2d

Part 1: Practical Exam

Task 1: Gateway With Wildcard TLS (10 Points)

Create the Gateway pruefung in infra (GatewayClass eg):

  • Listener http (port 80), only routes from its own namespace.
  • Listener https (port 443) for *.pruefung.gw.localtest.me, certificate pruefung-tls via the gateway-shim from lab-ca. Only routes from namespaces with gw-lab/team equal to shop or payments.
kubectl -n infra wait gateway/pruefung --for=condition=Programmed --timeout=5s
gw-ip.sh pruefung infra
kubectl -n infra get certificate pruefung-tls -o jsonpath='{.spec.dnsNames}{"\n"}'

Expected (the IP differs):

gateway.gateway.networking.k8s.io/pruefung condition met
172.18.0.9
["*.pruefung.gw.localtest.me"]

Task 2: Redirect HTTP to HTTPS (5 Points)

Every request on port 80 gets a 301 to the same URL with https. The route is called pruefung-redirect and lives in infra.

IP=$(gw-ip.sh pruefung infra)
curl -sI --resolve shop.pruefung.gw.localtest.me:80:$IP \
  'http://shop.pruefung.gw.localtest.me/x?y=1' | grep -iE '^(HTTP|location)'

Expected:

HTTP/1.1 301 Moved Permanently
location: https://shop.pruefung.gw.localtest.me/x?y=1

Task 3: Shop With Matching and Canary (15 Points)

HTTPRoute pruefung-shop in shop for shop.pruefung.gw.localtest.me:

  • /v2 and everything below → shop-v2.
  • Header x-canary: always → always shop-v2.
  • Everything else: 90% shop-v1, 10% shop-v2.

Task 4: API With Rewrite, Header and Timeout (10 Points)

HTTPRoute pruefung-api in shop for api.pruefung.gw.localtest.me: everything under /api goes to shop-v1 without the prefix (/api/produkte → /produkte), every response carries x-pruefung: bestanden, and the Gateway aborts after 2 s (504).

Task 5: Cross-Namespace With ReferenceGrant (10 Points)

The payments team publishes pay.pruefung.gw.localtest.me with an HTTPRoute pruefung-pay in the namespace payments that points to the Service shop-v1 in shop. The grant is called pruefung-pay and applies only to exactly this Service.

Task 6: Rate Limit (10 Points)

BackendTrafficPolicy pruefung-limit on pruefung-api: at most 5 requests per second (local, without Redis). The policy must report Accepted=True.

Task 7: Basic Auth for /admin (10 Points)

https://shop.pruefung.gw.localtest.me/admin only with user pruefer and password lab-pruef-1234 (lab value). A separate route pruefung-admin in shop to shop-v1, SecurityPolicy pruefung-admin, Secret pruefung-users.

Task 8: Repair a Broken Route (15 Points, Mandatory)

A colleague created this route. https://kaputt.pruefung.gw.localtest.me/ should deliver shop-v1. Repair only the route; the Gateway and Service stay as they are. There are three errors.

cat > kaputt.yaml <<'EOF'
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: pruefung-kaputt
  namespace: shop
spec:
  parentRefs:
    - name: pruefung
      namespace: infra
      sectionName: https-shop
  hostnames:
    - kaputt.gw.localtest.me
  rules:
    - backendRefs:
        - name: shop-v1
          port: 8080
EOF
kubectl apply -f kaputt.yaml
conds.sh httproute pruefung-kaputt shop

Expected (shortened, may differ depending on the version):

httproute.gateway.networking.k8s.io/pruefung-kaputt created
parent infra/pruefung/https-shop:
  Accepted=False (NoMatchingParent): No listeners match this parent ref
  ...

For each error, note the condition by which you recognized it.

Task 9: Two Envoy Replicas (5 Points)

The Gateway pruefung runs with two Envoy pods. Use an EnvoyProxy object pruefung-proxy in infra for this, not kubectl scale.

Task 10: Migrate an Ingress (10 Points)

Translate this Ingress with ingress2gateway (offline) and attach the result to infra/pruefung, listener https. https://legacy.pruefung.gw.localtest.me/alt/x must arrive at shop-v2 as /x. The route is called pruefung-legacy. You do not apply the Ingress itself.

cat > legacy-ingress.yaml <<'EOF'
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: legacy
  namespace: shop
  annotations:
    nginx.ingress.kubernetes.io/use-regex: "true"
    nginx.ingress.kubernetes.io/rewrite-target: /$2
spec:
  ingressClassName: nginx
  rules:
    - host: legacy.pruefung.gw.localtest.me
      http:
        paths:
          - path: /alt(/|$)(.*)
            pathType: ImplementationSpecific
            backend:
              service:
                name: shop-v2
                port:
                  number: 80
EOF
ingress2gateway print --providers=ingress-nginx --input-file legacy-ingress.yaml \
  -A --no-color > legacy-gw.yaml 2> legacy-warnungen.txt
grep -c 'kind: HTTPRoute' legacy-gw.yaml

Expected:

1

Check Script

cat > ~/gw-lab/pruefung/pruefung-check.sh <<'EOF'
#!/usr/bin/env bash
# Appendix B: check script – counts points per task
# shellcheck disable=SC2016  # check commands are in '…' and run via eval
set -uo pipefail
CA="${CA:-$HOME/gw-lab/tag04/lab-ca.crt}"
IP=$(gw-ip.sh pruefung infra) || exit 1
D=pruefung.gw.localtest.me
total=0; a8=0
pt() {  # pt <points> <text> <check command>
  if eval "$3" >/dev/null 2>&1; then
    total=$((total + $1)); printf '✔ %2s  %s\n' "$1" "$2"; return 0
  fi
  printf '✘  0  %s\n' "$2"; return 1
}
c() {  # c <host> [path] [curl options ...]: HTTPS against the exam Gateway
  curl -s --max-time 5 --cacert "$CA" --resolve "$1.$D:443:$IP" "${@:3}" \
    "https://$1.$D${2:-/}"
}
svc() { c "$@" | jq -r .service; }
st() { c "$@" -o /dev/null -w '%{http_code}'; }
anteil_v2() {  # 200 requests, shop-v2 must be between 8 and 40
  local n; n=$(for _ in $(seq 200); do svc shop; done | grep -c shop-v2)
  [ "$n" -ge 8 ] && [ "$n" -le 40 ]
}
ok_conds() {  # all conditions of all parents True (except RouteRulesOverlap)
  kubectl -n "$1" get httproute "$2" -o json | jq -e '[.status.parents[].conditions[]
    | select(.type != "RouteRulesOverlap" and .status != "True")] | length == 0'
}
echo "== A1"
pt 5 "Gateway pruefung Programmed" \
  'kubectl -n infra wait gateway/pruefung --for=condition=Programmed --timeout=5s'
pt 5 "Wildcard-Zertifikat aus lab-ca" 'openssl s_client -connect "$IP:443" \
  -servername "shop.$D" </dev/null 2>/dev/null | openssl x509 -noout -text \
  | grep -q "DNS:\*\.$D"'
echo "== A2"
pt 5 "HTTP -> 301 HTTPS" 'curl -sI --resolve "shop.$D:80:$IP" "http://shop.$D/x" \
  | grep -qi "^location: https://shop.$D/x"'
echo "== A3"
pt 5 "/v2 -> shop-v2" '[ "$(svc shop /v2/a)" = shop-v2 ]'
pt 5 "x-canary: always -> shop-v2" '[ "$(svc shop / -H "x-canary: always")" = shop-v2 ]'
pt 5 "Rest 90/10" 'anteil_v2'
echo "== A4"
pt 4 "Rewrite /api/produkte -> /produkte" \
  '[ "$(c api /api/produkte | jq -r .path)" = /produkte ]'
pt 3 "Header x-pruefung" \
  'c api /api/ -D - -o /dev/null | grep -qi "^x-pruefung: bestanden"'
pt 3 "Timeout 504" '[ "$(st api "/api/x?delay=3s")" = 504 ]'
echo "== A5"
pt 3 "ReferenceGrant auf genau shop-v1" 'kubectl -n shop get referencegrant \
  pruefung-pay -o json | jq -e "[.spec.to[]
  | select(.kind == \"Service\" and .name == \"shop-v1\")] | length == 1"'
pt 7 "pay -> shop-v1 aus payments" '[ "$(svc pay)" = shop-v1 ] \
  && kubectl -n payments get httproute pruefung-pay'
echo "== A6"
pt 5 "Policy pruefung-limit Accepted" 'kubectl -n shop get backendtrafficpolicy \
  pruefung-limit -o json | jq -e ".status.ancestors[0].conditions[]
  | select(.type == \"Accepted\" and .status == \"True\")"'
pt 5 "429 bei 30 schnellen Requests" '[ "$(for _ in $(seq 30); do st api /api/;
  echo; done | grep -c 429)" -gt 0 ]'
echo "== A7"
pt 5 "/admin ohne Login 401" '[ "$(st shop /admin)" = 401 ]'
pt 5 "/admin mit Login 200" '[ "$(st shop /admin -u pruefer:lab-pruef-1234)" = 200 ]'
echo "== A8 (Pflicht)"
pt 10 "kaputt -> shop-v1" '[ "$(svc kaputt)" = shop-v1 ]' && a8=$((a8 + 1))
pt 5 "pruefung-kaputt ohne False" 'ok_conds shop pruefung-kaputt' && a8=$((a8 + 1))
echo "== A9"
pt 5 "2 Envoy-Replikas über EnvoyProxy" 'kubectl -n infra get envoyproxy \
  pruefung-proxy && [ "$(kubectl -n envoy-gateway-system get deploy \
  -l gateway.envoyproxy.io/owning-gateway-name=pruefung \
  -o jsonpath="{.items[0].status.readyReplicas}")" = 2 ]'
echo "== A10"
pt 10 "legacy /alt/x -> /x bei shop-v2" 'kubectl -n shop get httproute \
  pruefung-legacy && [ "$(c legacy /alt/x | jq -r ".path + \" \" + .service")" \
  = "/x shop-v2" ]'
echo "== Ergebnis: $total von 100 Punkten"
if [ "$total" -ge 70 ] && [ "$a8" -eq 2 ]; then echo "✔ BESTANDEN"
else echo "✘ NICHT BESTANDEN (70 Punkte und Aufgabe 8 nötig)"; fi
EOF
chmod +x ~/gw-lab/pruefung/pruefung-check.sh
~/gw-lab/pruefung/pruefung-check.sh

Expected with a complete solution (shortened, may differ depending on the version):

== A1
✔  5  Gateway pruefung Programmed
✔  5  Wildcard-Zertifikat aus lab-ca
== A2
✔  5  HTTP -> 301 HTTPS
...
== A10
✔ 10  legacy /alt/x -> /x bei shop-v2
== Ergebnis: 100 von 100 Punkten
✔ BESTANDEN

The script is idempotent – run it after every task. The A6 check deliberately sends more than five requests per second; right afterwards, individual A4 checks may briefly see 429. In that case just wait a second and run it again.

Points

Task Topic Points achieved
1 Gateway with wildcard TLS 10
2 HTTP → HTTPS 5
3 Matching and canary 15
4 Rewrite, header, timeout 10
5 ReferenceGrant 10
6 Rate limit 10
7 Basic auth 10
8 Troubleshooting (mandatory) 15
9 EnvoyProxy 5
10 Ingress migration 10
Total 100

Clean up after the exam:

kubectl -n infra delete gateway pruefung
kubectl -n infra delete httproute pruefung-redirect
kubectl -n infra delete envoyproxy pruefung-proxy
kubectl -n infra delete secret pruefung-tls
kubectl -n shop delete httproute pruefung-shop pruefung-api pruefung-admin \
  pruefung-kaputt pruefung-legacy
kubectl -n shop delete backendtrafficpolicy pruefung-limit
kubectl -n shop delete securitypolicy pruefung-admin
kubectl -n shop delete secret pruefung-users
kubectl -n shop delete referencegrant pruefung-pay
kubectl -n payments delete httproute pruefung-pay

Part 2: Short Questions

1 point each, passing threshold 18 of 25. Answer in one sentence.

  1. Which three roles does the Gateway API separate, and which resource belongs to which?
  2. What does a GatewayClass with Accepted=Unknown and reason Pending mean?
  3. How do Accepted and Programmed differ on the Gateway?
  4. How are several entries under matches combined, and the fields within one entry?
  5. Does PathPrefix /v2 match /v2x? Why (not)?
  6. Two equally specific rules from two routes claim the same path. Who wins?
  7. A route has ResolvedRefs=False (BackendNotFound). What does the client see?
  8. Where do you create a ReferenceGrant when a route in shop points to a Service in payments?
  9. Which fields does a listener need for the cert-manager gateway-shim to issue a certificate?
  10. What distinguishes tls.mode: Terminate from Passthrough, and which route kind belongs to passthrough?
  11. What is validation.hostname in a BackendTLSPolicy for?
  12. What does weight: 0 mean, and why do you leave a backend in the route like that?
  13. Why is a canary with weights not a health check?
  14. What exactly does retry.attempts: 3 count?
  15. What distinguishes timeouts.request from timeouts.backendRequest?
  16. Which Envoy Gateway policy do you use for a rate limit, which for JWT, which for client timeouts?
  17. Why does a local rate limit of 10/s let more than 10 requests through with two Envoy pods?
  18. What happens to requests by default when Redis fails for the global rate limit?
  19. Why does Cilium L2 need externalTrafficPolicy: Cluster on the Envoy Service?
  20. What does the ValidatingAdmissionPolicy safe-upgrades.gateway.networking.k8s.io block, and what not?
  21. In which order do you upgrade Envoy Gateway, and why doesn’t helm upgrade do this on its own?
  22. What is the Envoy cluster of the first rule of the HTTPRoute shop/shop called, and what do you need the name for?
  23. What do the response flags UH and RLSE mean?
  24. Name two annotations that ingress2gateway does not translate or translates incorrectly.
  25. What is an InferencePool, and who picks the pod for a request within it?

Answer Key

  1. Infrastructure provider/cluster operator: GatewayClass; platform team: Gateway; app teams: Routes (and ListenerSets).
  2. No controller has claimed the class – the controllerName matches no installed implementation.
  3. Accepted: the spec is valid; Programmed: the data plane is configured and has an address.
  4. Multiple matches entries are OR, conditions within one entry are AND.
  5. No: PathPrefix compares whole path segments, /v2x is a different segment.
  6. The older route (creationTimestamp), on a tie alphabetically by namespace/name; Envoy Gateway reports RouteRulesOverlap.
  7. HTTP 500 for the affected rule – the route is accepted, its backend is not.
  8. In the target namespace payments, from HTTPRoute from shop, to Service (ideally with name).
  9. protocol: HTTPS, hostname, tls.mode: Terminate and certificateRefs (name = Secret) plus the issuer annotation on the Gateway.
  10. Terminate decrypts at the Gateway (L7 possible), passthrough passes TLS through unchanged by SNI – with TLSRoute.
  11. Envoy sends it as SNI to the backend and verifies the backend certificate against it.
  12. The backend gets no traffic, but stays in place for a rollback by swapping numbers.
  13. Weights also distribute to backends without endpoints – their share ends in 503.
  14. Three retries, so up to four attempts in total.
  15. request limits the total time until the response to the client including retries, backendRequest each individual attempt.
  16. BackendTrafficPolicy (rate limit), SecurityPolicy (JWT), ClientTrafficPolicy (client timeouts).
  17. Each Envoy pod has its own token bucket.
  18. They pass through without a limit (fail-open); only with rateLimit.failClosed: true do you get 500.
  19. Otherwise the IP may be announced on a node where no Envoy pod runs, and the traffic is lost.
  20. It blocks CRD bundles below v1.5 and Experimental over Standard; it lets Standard over Experimental through.
  21. Release notes, CRDs (helm template … | kubectl apply --server-side), controller, proxies – Helm only installs CRDs on the first install.
  22. httproute/shop/shop/rule/0 – the label envoy_cluster_name is the route key in PromQL and in the access log.
  23. UH: no healthy upstream (no endpoints); RLSE: rate limit service unreachable.
  24. For example rewrite-target with capture groups (missing), proxy-read-timeout (different value), snippets, auth-url.
  25. A backend object (inference.networking.k8s.io/v1) for model server pods; the Endpoint Picker (EPP) picks the pod via ext_proc.

Part 3: Self-Assessment

Topic I can do it done in the lab
Resource model, reading conditions (Day 1) [ ] [ ]
Gateway, listeners, matching (Day 2) [ ] [ ]
Filters, timeouts, retries (Day 3) [ ] [ ]
TLS, cert-manager, passthrough, BackendTLSPolicy (Day 4) [ ] [ ]
Canary and blue-green (Day 5) [ ] [ ]
GRPCRoute, TCPRoute, UDPRoute (Day 6) [ ] [ ]
allowedRoutes, ReferenceGrant, ListenerSet, RBAC (Day 7) [ ] [ ]
RKE2 HA on Proxmox with Cilium (Day 8) [ ] [ ]
LB IPAM, L2, two GatewayClasses, DNS (Day 9) [ ] [ ]
Client, backend, security policies (Day 10) [ ] [ ]
Metrics, access logs, tracing, status alerts (Day 11) [ ] [ ]
HPA, PDB, drain, upgrades, certificate rotation (Day 12) [ ] [ ]
Ingress migration with parallel operation (Day 13) [ ] [ ]
Disaster drills under time pressure (Day 14) [ ] [ ]

Any row without two checkmarks: work through that chapter once more.