Mapping hops before blaming the database
When latency rises, the database is an easy suspect. Sometimes it is guilty. Often the wait sits in a load balancer queue, a broker retry loop, or a storage volume that hit IOPS ceilings while the query plan looked fine.
Flow visibility mapping starts with a blunt inventory: every hop from client entry to the last write, with a human owner beside each box. Missing owners are findings. So are hops that exist in production but not in the architecture wiki.
Spend half a day on that map before rewriting indexes. You may still touch the database — but you will know whether it is the bottleneck or the scapegoat.