Seeing what the gateway sees

Key points

• Every request passes through Kong, so it is the natural place to measure latency, status and volume.
• The prometheus plugin exposes per-route metrics with no change to any service.
• Correlation-id generates a request id and passes it upstream, so one id follows a call across services.
• Logging plugins run in the log phase, after the client has its response - so they add no latency.
• Watch upstream latency and Kong latency separately: they point at completely different problems.

Example

plugins:
  - name: prometheus
    config: { status_code_metrics: true, latency_metrics: true }

  - name: correlation-id
    config:
      header_name: X-Request-Id
      generator: uuid
      echo_downstream: true

  - name: http-log            # runs in the log phase - off the request path
    config: { http_endpoint: http://logstash:8080 }

Split gateway latency from upstream latency, or you will scale the wrong thing.