Load Testing Services

Bottleneck identification during a load test uses correlated monitoring across all application layers simultaneously. Infrastructure layer: CloudWatch metrics for ECS CPU utilisation, RDS CPU and connections, ElastiCache memory, ALB 5xx rate watch which metric saturates first as load increases. Application layer: Node.js inspector or clinic.js heap profiling under load is CPU saturating in the application (event loop blocking) or are most threads waiting for I/O? Database layer: pg_stat_statements query durations during the load test do query times increase with load (indicates lock contention or resource exhaustion) or stay constant (indicates the query is efficient at any load)? Connection pool: PgBouncer stats (pool wait time, queue length) if applications are waiting for a database connection, PgBouncer pool size is the bottleneck. The bottleneck is almost always the first resource that saturates increasing load beyond that point degrades everything downstream. ClickMasters instruments all layers before running a load test to ensure bottleneck identification is immediate.

Auto-scaling validation ensures your infrastructure can handle traffic spikes without manual intervention. The key metrics: trigger threshold (what CloudWatch alarm triggers the scale-out? at what CPU or request count?), scale-out timing (how long from trigger to new instances receiving traffic? If auto-scaling takes 3 minutes but the spike lasts 2 minutes, it doesn't help), scale-under-load (are new instances healthy and processing traffic before existing instances are overwhelmed?), database response under scale-out (does adding application instances cause database connection exhaustion? if yes, no amount of application scaling helps), scale-in validation (after load decreases, do instances scale in correctly? no stuck instances, no premature scale-in causing second degradation). ClickMasters validates auto-scaling configuration as part of the load test including measuring the scale-out timing and validating that the database layer can handle the increased connection demand.