Coherencia de datos entre cachés

When you have multiple layers of caching, like a browser cache, a CDN, an edge cache, and an application cache (e.g., Redis), data consistency becomes a big challenge.

Users expect to see the most up-to-date information. If your product price changes in the database, but an old price is served from a CDN, that's a problem!

Imagine your data journey:

• Origin Database: The source of truth.
• Application Cache (Redis): Stores frequently accessed data for your app.
• Edge Cache: Caches dynamic content closer to users.
• CDN: Caches static assets globally.
• Browser Cache: Your user's local cache.

Each layer can hold a copy of the data. How do you ensure they all reflect changes from the origin?

Stale data is information in a cache that is no longer current because the original data source has been updated.

If a blog post title is updated in your database, but the CDN or browser still serves the old title, that's stale data. It leads to confusion and a poor user experience.

Time-To-Live (TTL) is a simple way to manage cache freshness. Each cache entry is given a lifespan. After this time, it's considered stale and must be re-fetched.

While useful, TTL alone isn't perfect for immediate consistency across *all* layers. If data changes before its TTL expires, caches will still hold stale data until the TTL runs out.

To achieve better consistency, especially for critical data, we need proactive invalidation. This means actively telling caches to remove or refresh specific data when the origin changes.

Instead of waiting for TTL, we trigger an invalidation event immediately after data is updated.

A powerful technique for multi-layer consistency is event-driven invalidation. When data changes at the origin, an event is published to a messaging system (like Redis Pub/Sub, Kafka, etc.).

Different caching layers (application, edge) can subscribe to these events and invalidate their local copies instantly.

Here's a conceptual example of how an application might signal an invalidation for a specific item after it's updated in the database.

This signal would then be picked up by other services or caching layers to clear their data.

Another strategy is cache tagging or versioning. Instead of just invalidating, you change the identifier of the content when it updates.

• URL Versioning: /image.jpg?v=123 becomes /image.jpg?v=124
• ETags: HTTP header ETag: "abcdef" changes to a new value.

This tells CDNs and browsers that it's a completely new resource, forcing a fresh fetch.

For complex multi-layer systems, you often need an invalidation orchestration layer. This is a dedicated service or logic that understands all your cache layers.

When an update occurs, this orchestrator receives the event and then intelligently triggers purges on your CDN, invalidates specific keys in Redis, and perhaps signals edge caches to refresh.

Maintaining data consistency across multiple caching layers is vital for a good user experience. We explored several strategies:

• Understanding the limitations of simple TTL.
• Implementing proactive, event-driven invalidation across layers.
• Utilizing cache tagging or URL versioning to force fresh content.
• Considering an invalidation orchestration layer for complex systems.

By combining these techniques, you can ensure your users always see the freshest data, no matter how many caches are involved!