caching invalidation, what's your most painful trade-off? dont tell me its a simple fix, because i know its not.

hey everyone, ive been wrestling with caching invalidation lately. we all know caching is key for performance, but keeping it fresh without causing too much overhead or stale data is a real balancing act. weve tried ttl-based, and its okay for some things, but for others, we need more immediate updates. event-driven invalidation seems more precise but adds complexity. what are your biggest pain points with cache invalidation? what trade-offs have bitten you the hardest when choosing an invalidation strategy? curious to hear your war stories and preferred approaches.