The freeze that started three layers away
The ticketing system froze every peak day, and everyone looked at the ticketing system. The load balancer logs led to dashboard code calling an internal API on every login, and behind it, an API that could not scale. Symptom in one system, cause in another, evidence in a third. Code rewritten, API scaled, season saved.
SYMPTOM → LOGS → CODE → CAUSE
When adding servers made nothing faster
A legacy application autoscaled onto a shared FSx cache and latency never moved, while AWS's dashboards called the storage healthy. The bottleneck was in metadata operations, visible only when CloudWatch agent metrics were read against a load-test baseline. The cache moved to NVMe instance volumes with stickiness, and the system held 170,000 requests per minute, 10,000 concurrent users in a 15-minute window.
170,000 REQ/MIN · 10,000 CONCURRENT USERS