What's your go-to strategy for cache invalidation in a distributed system?

We're having a few problems with old data showing up across our services, and it looks like just expiring stuff by time isn't really doing the trick anymore. Our users are super active, and the data changes a lot, which means we're seeing some real inconsistencies. So, I'm wondering, how are other folks managing to clear their caches in a truly distributed setup? Are you mostly going with event-driven invalidation using pub/sub, or maybe more involved consistent hashing, or something completely different? What kind of good and bad points have you noticed with these different ways, especially when it comes to how complicated they are, how fast they run, and if the data stays accurate?