Design a cache invalidation strategy for a global e-commerce platform where product inventory is distributed across multiple database shards and cached in Redis clusters located in US-East, EU-West, and AP-South. The system must handle high write throughput while minimizing the probability of users reading stale inventory data. Describe the trade-offs between write-through, write-back, and write-around patterns in this context, and propose a mechanism (such as versioning or pub/sub) to handle invalidation messages across regions with network latency considerations.