Design a global certificate issuance and rotation service

Design the service that keeps 5 million customer domains on a global edge supplied with valid public TLS certificates as maximum lifetimes fall towards 47 days: ACME with more than one CA, domain validation that works for delegated customers, renewal guided by ARI, keys distributed to 300 locations, and the ability to replace every certificate within 24 hours when a CA has to revoke.

● Hard🔒 PROBy ·Design problem·15 min read·Advanced → Staff
ACMEcertificate rotationARIcertificate transparencyCAAmass revocation

01Understand the problem and size it

✓What you will learn
  • Plan for the CA/Browser Forum lifetime cuts: 200 days from March 2026, 100 days from March 2027, 47 days from March 2029
  • Issue at scale with ACME (RFC 8555) across several CAs, with validation methods that suit delegated domains
  • Renew on ARI (RFC 9773) guidance and survive a mass revocation in 24 hours
  • Distribute private keys safely to 300 locations and verify what is actually served
◎Interview radar

What they probe: whether you know where certificate lifetimes are heading and what that does to volume, how you validate domains you do not control DNS for, how you avoid depending on one CA, what you do when a CA must revoke millions of certificates within days, and how private keys reach the edge. Where candidates slip: one CA, renewing on a fixed 30-days-before-expiry timer, ignoring CAA records, assuming OCSP-based revocation checking still matters at public CAs, no independent expiry monitoring.

A platform that serves customers' websites (a CDN, a hosting provider, a SaaS with custom domains) must hold a valid public certificate for every customer hostname. Certificate lifetimes are shrinking by rule: in April 2025 the CA/Browser Forum approved ballot SC-081, which cuts the maximum TLS certificate validity to 200 days from 15 March 2026, 100 days from 15 March 2027 and 47 days from 15 March 2029, with domain validation reuse periods shrinking alongside. Manual renewal is already impossible at this scale; soon even slow automation will be.

Functional requirements

  1. Issue and renew certificates for every customer hostname, from more than one public CA.
  2. Validate domain control whether the customer points DNS at us (CNAME) or delegates validation.
  3. Distribute certificates and keys to every edge location; serve the right one by SNI.
  4. Replace any set of certificates (one CA's, one customer's, all) on demand.

Non-functional requirements

  1. 5 M hostnames, 300 PoPs.
  2. No customer outage from expiry; renewal starts with at least a third of the lifetime remaining.
  3. Full reissue of all certificates from one CA within 24 hours.
  4. Private keys never stored or sent in plaintext.
Sizingtxt
hostnames                5,000,000 (one certificate each for simplicity)
today (90-day certs, renew at 60 days)    5 M / 60 days   ≈ 83,000 / day ≈ 1 / s
2029 (47-day max, renew every ~30 days)  5 M / 30 days   ≈ 167,000 / day ≈ 2 / s
mass reissue in 24 h     5 M / 86,400 s ≈ 58 / s sustained, all validations included
  -> far above routine; CA per-account rate limits must be agreed in advance and the load
     spread across CAs
certificate + chain + wrapped key  ~5 kB -> 5 M x 5 kB = 25 GB for the full set
edge                     PoPs keep hot hostnames in memory, fetch the rest on first SNI
▭On the whiteboard · Scope3 points
01
FR
issue + renew from several CAs | validation for CNAME and delegated customers | distribute by SNI | targeted reissue
02
NFR
5 M hostnames, 300 PoPs | no expiry outages | 24 h full reissue per CA | keys never in plaintext
03
out
internal service mTLS (a private PKI), customer-uploaded certificates

This is an advanced module

Design problems: global security control planes is part of the Pro pass. 4 more sections — the deep dives, the final design, follow-ups and the test — unlock with it.

Already a cyart.io or studio.cyart.io member? Your pass code is in your Studio account.