API Gateway Pipeline Lab (Interactive)
Route valid, spoofed, flooded, and junk requests through togglable TLS, auth, rate-limit, tracing, and canary stages. Toggle each perimeter stage, bypass the gateway entirely, or smuggle tax logic into the edge and watch headers, shed rates, hop latency, and event-loop capacity respond.
API Gateway Edge Pipeline
Send hostile and ordinary traffic through the perimeter stages and watch what each layer — or its absence — decides.
GET /v1/orders · valid JWT
HTTP 200
headers forwarded over VPC:
x-user-id: usr_9981
x-tenant-id: org_4401
traceparent: 00-0000000036147e2f…
edge hop +2.2 ms gateway tier capacity 62,000 rps
One edge tier verifies credentials once, strips forged identity headers, sheds floods with 429 + Retry-After, and emits traceparent spans — services stay private inside the VPC behind multi-AZ replicas.
How It Works Under the Hood
With dozens of microservices, direct client access means every service exposes a public IP, manages TLS, and re-implements JWT parsing, CORS, and rate limiting. The gateway centralizes the perimeter: it terminates TLS, validates tokens once, strips forged inbound identity headers, injects trusted X-User-Id and traceparent, sheds floods against Redis token buckets before backend compute, and canary-splits traffic by hashing. The one cardinal rule is staying dumb — embedding domain logic in Envoy event loops turns the shared edge into a monolithic bottleneck that degrades every unrelated endpoint.
Core Architectural Principles
- Edge JWT validation plus header sanitization lets internal services trust injected identity.
- Per-stage cost accounting: TLS, auth, rate limit, tracing, and canary routing each add sub-millisecond hops.
- Business logic in the gateway blocks async event loops, cratering throughput mesh-wide.
Describe the perimeter pattern concretely: gateway verifies tokens and injects X-User-Id plus correlation IDs, sheds load with 429 and Retry-After, routes canaries, and holds zero domain logic. Then cover its own availability: multi-AZ auto-scaling replicas behind an L4 NLB, because a lone gateway is a single point of failure for every service behind it.
One extra 0.5-2ms hop and SPOF risk versus not duplicating security plumbing across every microservice.