Cache invalidation

Cache invalidation connects writes to stale reads through TTL expiry, explicit delete, versioned keys, tag-based purge, or event-driven invalidation. There is no universal best strategy — combine short TTL as safety net with proactive invalidation on known dependencies.

Version keys include updated_at or content hash so new writes naturally miss old entries. Tag queues map related keys for bulk purge when a category changes.

On interviews: explain invalidation for "user updates profile photo" across CDN, Redis, and client cache layers.

Common pitfalls: invalidating too little (stale UI) or too much (thundering herd on origin); race between write and invalidation; forgotten edge caches.

The trade-off is flexibility versus complexity—know when the simpler path is enough.

Checklist:

• Map each write to affected cache layers.
• Use TTL plus proactive invalidation.
• Prefer versioned keys where possible.
• Test invalidation under concurrent updates.