Would you use Redis pub/sub or a durable queue for cache invalidation events?

In production I would combine mechanisms when useful: a durable event for reliable processing and short TTLs or versioned keys as a safety net. I would monitor subscriber errors, invalidation lag, dead letters, and stale-read incidents, then verify that a restarted instance catches up correctly. Redis pub/sub is attractive because it is simple, but I would use it only where missed messages create bounded staleness rather than data corruption or security exposure. For critical invalidation I would favor a durable topic, replayable event, or version bump that makes old cache entries unreachable.