Fix it: the slow database
Statement
You inherit a running system: a load balancer, three server groups and a tier “m” database. Fine by day, but in the evening at 2000 requests/s (85 % reads) users complain about timeouts. Change the diagram so it meets the SLO again. There are several fixes: scale the database vertically, add a cache or a replica. The budget is $2000; the cheaper, the more stars.
Constraints
- p95 ≤ 250 ms, errors ≤ 0.1 %, budget $2000/month, ≤ 5 instances per node.