How do I identify the slow part of an API?

Definition:

API latency is the total time a request spends waiting and doing work across the system. That time can include load-balancer delay, connection-pool wait, application CPU, database work and external calls. Breaking it into stages turns “the API is slow” into a testable explanation.

Key points:

• Measure request time, database time, external calls and queue/pool waits separately.

• Check p95/p99 latency too; averages can hide slow requests.

• Compare the same traffic and workload before and after a fix.

• Distinguish throughput from latency: many completed requests per second can coexist with a badly delayed minority of requests.

Interview point:

• Use percentiles and traces to find the affected users and slow stage rather than relying only on average response time.