Course contentsShow
← System Design
Lesson 949 of 1,91924. Authentication and Authorization at ScaleFree lesson

Clock Skew and Token Validation

How clock differences between servers affect token expiration validation and mitigation strategies.

Clock Skew and Token Validation

What you'll learn: How time differences between servers can break token expiration checks and how to handle them gracefully.

The Time Synchronization Problem

When you issue a token with an expiration timestamp (like a JWT with an exp claim), you're assuming all servers share the same clock. In reality, distributed systems have clock skew—small time differences between machines. A token issued by Server A at "12:00:00" might be validated by Server B whose clock reads "11:59:55". If the token's nbf (not-before) claim says "12:00:00", Server B rejects it as "not yet valid," even though it's perfectly legitimate.

Think of it like crossing time zones: you board a plane at 3 PM in New York with a ticket valid "after 3 PM," but the system in Chicago (still 2 PM local time) might reject it if it naively compares local clocks.

Common Issues

  • Premature rejection: Tokens rejected as expired when they're still valid
  • Delayed expiration: Tokens accepted when they should be expired
  • nbf validation failures: Tokens marked "not before" a future time

Mitigation Strategies

  1. Clock skew tolerance (leeway): Accept tokens within a grace period (e.g., ±5 minutes). Most JWT libraries support this via a clockTolerance parameter.
  2. NTP synchronization: Keep all servers synchronized using Network Time Protocol, reducing skew to milliseconds.
  3. Short-lived tokens: Use 5-15 minute expiration times, making clock skew less impactful relative to token lifetime.
  4. Centralized validation: Route token checks through a single service with one authoritative clock.

Key Takeaway: Clock skew is inevitable in distributed systems; always implement leeway tolerances when validating time-based token claims and keep servers NTP-synchronized to minimize drift.