Recovering From Feed Cache Loss

A Redis node holding feed caches for a shard of 2 million users fails unrecoverably and its replica had fallen behind, so the feeds for those users are effectively gone. On-call wants to know: is this a data-loss incident?

1. Is it a data-loss incident? Explain what is actually lost versus what can be recovered, and from where.
2. Design the recovery path for an affected user's next feed read. What does it do differently from the normal (warm-cache) path, and what is the user-facing effect?
3. If all 2 million users request their feed within the same minute after the incident is noticed (e.g. a mobile client retry storm), what could go wrong, and what would you do to prevent it?