Challenge 1: Sentinel or Cluster? — Possible Solution
====================================================================

SENTINEL is the right fit here.

The stated requirement is specifically about SURVIVING A FAILURE
without manual intervention — the dataset itself is explicitly said to
already fit comfortably on one instance, so there is no scaling
problem to solve at all. Sentinel is built exactly for this: it
monitors a primary and its replica(s), and automatically promotes a
replica to primary if the current one goes down, with no human needing
to intervene.

CLUSTER ISN'T RELEVANT HERE because Cluster solves a completely
different problem — splitting a dataset that has outgrown a single
node's capacity across multiple nodes. Since the dataset already fits
comfortably on one instance, there's nothing to shard; introducing
Cluster would add real operational complexity (hash slots, cross-slot
key restrictions, more nodes to manage) to solve a scaling problem
this team doesn't actually have, while still needing something like
Sentinel (or Cluster's own separate failure detection) layered on top
to actually satisfy the "survive a failure automatically" requirement
anyway.

The right match is: has a failure-tolerance need, no scaling need ->
Sentinel with plain primary-replica replication, not Cluster.