Not an alarm — this is the proactive end of the queue. The same class of read pressure that tripped the RDS CPU alarm lurks on the busiest read endpoints. Rather than wait for them to page, harden them first.
Each endpoint was made faster and more resilient under load — tightening the queries and response mapping and leaning on Redis where the same data is served repeatedly, so a burst of traffic doesn't translate into a database spike.
The hottest read paths are quicker and far less likely to be the source of the next alarm. This is the half of production engineering that never shows up as a ticket from a customer — the agent freeing human time to spend here is the point.