An attack surface is the set of exposed interfaces and entry points through which an unauthorized party could affect a system. In a service graph, inbound arrows from browsers or external webhooks mark the public API and security attack surface. Trust boundaries become visible when connections are classified as inbound, internal, or outbound. section 9.1
Map the surface by asking:
- Which nodes receive public traffic, and which gateway configuration defines their routes? section 9.1
- Which edges cross the internet, and which connect to vendor systems such as Stripe or Google? section 9.1
- Which components are custom code requiring inspection, rather than commodity or rented infrastructure? section 9.1
A recognizable web boundary
A browser or external webhook points into the web application, creating an inbound public edge. An Nginx configuration, Kubernetes Ingress manifest, or Kong configuration may define most routes and therefore much of the exposed surface. section 9.1
Table 9.1 A classification of architectural nodes and edges
|
Term
|
What it is
|
Example
|
Why it
matters
|
|
Node
|
A
long-running program or service
|
A web
server, Postgres, Stripe
|
The box
that can fail
|
|
Edge
|
A wire
between two nodes
|
A SQL
query, an HTTP call, a queued job
|
The wire
that times out
|
|
Run
|
Your code
on your machine
|
Their own
web app, their own worker
|
You can
edit anything
|
|
Rent
|
Vendor code
on your machine
|
Postgres,
Redis, RabbitMQ
|
Configure
only; no source edits
|
|
Call
|
Vendor code
on a vendor's machine
|
Stripe,
SendGrid, Twilio
|
Retry,
cache, or route around
|
|
Client
|
Your code
on a machine you don't operate
|
Browser
frontend, mobile app, desktop app
|
Ship
updates, no secrets, untrusted input
|
|
Sync edge
|
The user is
waiting on it
|
Checkout →
Stripe
|
The user
feels its latency directly
|
|
Async edge
|
The user
has already moved on
|
Queue →
send confirmation email
|
The user
never sees it; it can fail silently
|
|
Public edge
|
Crosses out
to the open internet
|
A webhook
from Stripe → you
|
The
security boundary; this is where attackers knock
|
|
Internal edge
|
Between two
of your own nodes
|
Web →
Postgres
|
A network
hop, but trusted; failure is the wire, not the user
|
Classify nodes and edges by ownership, location, direction, and whether users wait on the connection. section 9.1






