scalability-playbook

Start by profiling current metrics: request rate, p95 latency, CPU, memory, database query time, and cache hit rate. Map each symptom to a measurement, then assign trigger thresholds (warning and critical) so you know exactly when to act on each bottleneck.

Start with low-cost quick wins: add database indexes on frequently queried columns and introduce a Redis caching layer for repeated reads. These typically deliver 60-80% improvement for under $50/month before considering read replicas or sharding.

Use read replicas first for read-heavy workloads since they are cheaper and take about a week to deploy. Reserve sharding for when a single database exceeds storage or write capacity limits, as it costs more and requires months of effort.

Follow a phased plan: add indexes and caching in weeks 1-2, read replicas by week 4, horizontal app servers by week 6, and a CDN by week 8. Validate each step with load testing tools like hey at 10x concurrency before production rollout.

Key triggers include CPU above 70%, memory above 75%, database connections above 40, query p95 above 500ms, queue depth above 1000, and error rate above 1%. Each threshold maps to a specific action such as adding servers, replicas, indexes, or workers.