Scalability and Authentication at Twitch

Ideas you will meet

Distributed Systems

Consensus, time, failure and the fundamentals of scale-out.

Web Security

OWASP risks, XSS, CSRF, CORS and secure design.

What we already know about this

Our own lessons on these ideas. Read them here before you start, so it makes sense.

• Real-World Systems · 10 minSecuring a web APIAuthentication, authorisation, input handling and abuse protection: the layers that keep an API safe.Read hereClose

  Think in layers

  No single control secures an API. Defence comes from several layers, each assuming the others may fail.

  100%

  Figure SDA-RWS-14-1 · API security layers · © forcsguy.com

  1. Transport

  Use HTTPS everywhere, with modern TLS and HSTS. Never accept credentials over plain HTTP.

  2. Authentication: who is calling?

  • Users: OAuth 2.0 / OpenID Connect access tokens, or secure session cookies for browsers (see the earlier lessons).
  • Machines: API keys or client credentials, stored as hashes, scoped, and rotatable.
  • Send credentials in the Authorization header, never in URLs.
  • Keep token lifetimes short, support revocation, and protect refresh tokens carefully.
  • Add multi-factor authentication for sensitive accounts, and protect login endpoints with rate limits.

  3. Authorisation: what may they do?

  This is where most real API breaches happen.

  • Check on every request, on the server, both the action (role or scope) and the specific object. "Is this user logged in?" is not enough; ask "may this user read order 123?" (broken object-level authorisation is the top API risk).
  • Apply least privilege: tokens carry only the scopes they need.
  • Never rely on the client hiding buttons.
  • Do not expose sequential ids as a security measure. Use authorisation checks (random ids only help a little).

  4. Input and output

  • Validate every input against a schema: types, lengths, ranges, allowed values. Reject unknown fields (prevents mass assignment, where a caller sets fields like isAdmin).
  • Use parameterised queries to prevent SQL injection, and encode output to prevent injection into other systems.
  • Limit request size and file types, and scan uploads.
  • Return only the fields needed (avoid excessive data exposure), and paginate lists.
  • Return generic error messages to clients and keep detail in logs, so errors do not leak internals.

  100%

  Figure SDA-RWS-14-2 · Two ways a caller proves who it is · © forcsguy.com

  100%

  Figure SDA-RWS-14-3 · Signing each request with a shared secret (HMAC) · © forcsguy.com

  Image: Spanish Coches · CC BY 2.0 · Source · Wikimedia Commons: Peaje - Toll - Péage (5020183265)

  5. Abuse protection

  • Rate limiting per user, key and IP, with stricter limits on expensive or sensitive endpoints.
  • Quotas and timeouts; protect against huge queries (depth limits in GraphQL, page size caps).
  • Bot and credential-stuffing defences: lockouts, delays, CAPTCHAs, breached-password checks.

  6. Webhooks and service-to-service calls

  • Sign outgoing webhooks with an HMAC over the body and a timestamp. Receivers verify the signature in constant time, reject old timestamps and handle events idempotently.
  • Between services use mutual TLS or signed tokens, and network policies, instead of trusting "internal" traffic.

  7. Browsers: CORS and cookies

  • Configure CORS with an explicit allow-list of origins. Never reflect any origin with credentials enabled.
  • Cookies: HttpOnly, Secure, SameSite. Protect state-changing requests from CSRF.

  An example worth knowing: in 2021 a flaw in a popular Java logging library (Log4Shell) let attackers run their own code simply by getting a crafted string written to a log.

  100%

  Figure SDA-RWS-14-4 · How untrusted text became remote code execution · © forcsguy.com

  8. Secrets, data and visibility

  • Keep secrets in a secrets manager, not in code or images, and rotate them.
  • Encrypt sensitive data at rest, and minimise what you store.
  • Log authentication events and authorisation failures (never secrets), and alert on anomalies. Keep an inventory of your APIs so old versions and forgotten endpoints do not stay exposed.
  • Keep dependencies patched, and test: security tests, dependency scanning, and periodic penetration testing.

  More on this site

  Related reading

  • MDN: Cross-Origin Resource Sharing (CORS)MDN contributors · MozillaDocumentation
  • Introduction to JSON Web TokensAuth0 · jwt.ioDocumentation
  • RFC 6749: The OAuth 2.0 Authorization FrameworkIETFDocumentation
  • AWS Security BlogAmazon Web Services engineers · Amazon Web ServicesBlogs

  Show 6 moreShow fewer

  • OWASP Cheat Sheet SeriesOWASP · OWASP FoundationDocumentation
  • Google Security BlogGoogle engineers · GoogleBlogs
  • How we built rate limiting capable of scaling to millions of domainsCloudflare engineering · CloudflareCase Studies
  • OWASP Top 10OWASP · OWASP FoundationDocumentation
  • Rate LimitingKeycdn engineers · KeycdnCase Studies
  • Scalability and Authentication at TwitchTwitch engineers · TwitchCase Studies

  Open this lesson with its quiz and exercise