Cloud-Native API Gateway Practitioner (Course #cloud-native-gateway)

For platform engineers, integration architects, Kubernetes operators, API developers.

Ref#cloud-native-gateway
Duration3 days
Formats4 ways
Cert prep—
PrerequisitesNone

By the end, you can

By the end of this course, you will be able to design and operate a Gateway API estate on two independent data planes — role-separated route definitions, portable identity and rate-limit policy, and a GitOps reconciliation loop — so that changing gateway product becomes a platform decision rather than a rewrite.

Running case study

Saltmere Utilities exposes metering, outage and billing APIs to 14 retail partners from a single Kubernetes platform serving 4 regions. Ingress config is 240 annotations deep and only two engineers can safely change it, so a routing change takes 9 days of review. The platform team is standardising on Gateway API with a target of route changes merged by service teams in under 1 hour, no partner-visible downtime, and a data plane that can be replaced without rewriting the routes.

Programme every line is something you do

7 modules · 21 topics · in course-flow order

  1. Module 1

    Three roles, three resources

    Objective. Design the role separation Gateway API assumes and produce the RBAC split that lets service teams own their own routes.

    • Map GatewayClass, Gateway and HTTPRoute onto the 3 personas the specification names: infrastructure provider, cluster operator, application developer
    • Split RBAC so a service team merges a route without touching listeners, reducing the 9-day review to a single team approval
    • Model listener, hostname and route attachment rules, and prove which namespace may attach to which listener
  2. Module 2

    Routes that survive a rename

    Objective. Construct a route set whose matching, rewriting and traffic-splitting behaviour is specified rather than inherited from a controller default.

    • Express path, header and method matching precisely, and resolve precedence between 4 overlapping rules
    • Implement weighted traffic splitting for a canary at 5% and prove the split with request-level evidence
    • Move filters — rewrite, redirect, request-header mutation — out of annotations and into declared route filters
  3. Module 3

    Kong end to end

    Objective. Deploy the reference route set on Kong and record which behaviours require Kong-specific extension.

    • Install the controller and bind a GatewayClass, then reconcile the full route set with zero manual edits
    • Attach 3 policies — authentication, rate limiting, request transformation — and record which are portable and which are Kong-specific
    • Read the plugin execution order explicitly; priority changes between releases silently reorder a policy chain
  4. Module 4

    Traefik end to end

    Objective. Deploy the identical route set on Traefik and produce the divergence report between the two data planes.

    • Reconcile the same route set on a second data plane and count the resources that needed no change
    • Diff behaviour across the 2 implementations for matching precedence, header handling and error responses
    • Record every divergence as either a conformance gap or an implementation-specific extension, with the evidence for each
  5. Module 5

    Identity at the edge

    Objective. Implement OAuth 2.0 and OIDC verification at the gateway and prove the identity the backend actually receives.

    • Validate JWTs at the edge — issuer, audience, expiry, key rotation — and reject on all 4 without reaching the backend
    • Terminate and re-originate TLS for a mutual-TLS partner, keeping the client identity readable downstream
    • Diff the request before and after the gateway to prove no authorization header is dropped or remapped in transit
  6. Module 6

    Limits that hold

    Objective. Configure rate limiting whose enforcement is measured under partition rather than assumed from configuration.

    • Choose a counting strategy deliberately: local, cluster and shared-store counting differ by a factor of N nodes
    • Cut the counter store for 60 seconds under load and record what the limiter actually enforced during the outage
    • Compare the data-plane implementations of the same limit on Envoy and Kong, and document the accuracy each achieves
  7. Module 7

    Reconcile, observe, prove

    Objective. Operate the estate through GitOps and demonstrate its behaviour under a load profile matching production peak.

    • Reconcile declared against running configuration on every merge and fail the build on any drift
    • Emit RED metrics and distributed traces from the data plane, with a trace sampled on at least 1% of requests
    • Load-test to the 4-region peak profile and record p99 latency and error rate against a 400 ms target

References & Official Documentation every claim above has a source

Module 1

Module 2

Module 3

Module 4

Module 6

More in Architecture

Want This Course for Your Team?

We tailor every delivery to your use cases, data, and schedule.

Request a Quote