How do you invalidate a cache?

Start from the staleness budget. If marketing tolerates a price change appearing within a minute, TTL = 60 s with cache-aside is enough and simple. If a price must never be stale, the write path invalidates: update Postgres, then delete the cache key; the next read repopulates. Delete rather than set on write, because a set can race with a concurrent read that populates an older value. Keep a TTL anyway as a safety net for missed invalidations.

Stampedes: when a popular key expires, hundreds of requests miss simultaneously and all hit Postgres. Mitigations: a short lock so one request recomputes and the rest wait or serve the stale value; probabilistic early recompute so the key refreshes shortly before expiry; TTL jitter so keys created together do not expire together — the midnight expiry of every key set at midnight is the classic outage. Negative caching (cache "not found" briefly) stops a missing product id from hammering the database.

Hot keys: a single product on the homepage can saturate one Redis shard; an in-process cache with a few seconds TTL in front of Redis handles that, at the cost of per-instance staleness.