Optimizing cache invalidation: pub/sub vs. write-through for distributed caches?

Hey everyone, I'm working on a new service that heavily relies on caching, and I'm trying to decide on the best cache invalidation strategy for a distributed cache (think Redis cluster or Memcached). I'm weighing the pros and cons of using a pub/sub mechanism for cache invalidation messages against a write-through approach where writes update both the database and invalidate/update the cache directly. Pub/sub seems more scalable for read-heavy scenarios, but introduces potential latency and complexity with message ordering. Write-through feels simpler to implement initially, but I worry about the write amplification and potential for contention with frequent updates. What have your experiences been with these approaches in production? Any specific scenarios where one clearly outperforms the other? Are there other patterns I should consider for keeping distributed caches fresh?