Cache invalidation strategies: our journey from time-based TTL to event-driven

we've been evolving our cache invalidation strategies over the years. initially, we started with simple time-based ttls, usually 5 minutes, but that led to users seeing stale data. reducing the ttl to 30 seconds helped with freshness but resulted in a 10x increase in database load due to more frequent cache misses. our current approach is event-driven invalidation: when data changes in the primary source, we publish an event that invalidates the corresponding cache entry. this works pretty well but requires careful event design and ensuring all producers publish correctly. we're now considering change data capture (cdc) to automatically invalidate caches based on database changes, which sounds promising for reducing developer toil. what are your current go-to cache invalidation strategies, especially for critical data, and what challenges have you overcome with them?