What is the best way to define a Service Level Indicator (SLI) for scaling an NGINX-based application deployed in GKE?

The best way to define a Service Level Indicator (SLI) for scaling an NGINX-based application deployed in Google Kubernetes Engine (GKE) is to install the Cloud custom metrics adapter and configure a horizontal pod autoscaler to use the number of requests provided by the Google Cloud Load Balancer (GCLB).

This approach is effective because it allows dynamic scaling of application pods based on actual user demand. By leveraging the number of incoming requests being handled by the load balancer, the autoscaler can make more informed decisions about scaling up or down the number of pods. This ensures that the application remains responsive and available under varying load conditions, directly aligning with performance metrics that are critical for maintaining high availability and service quality.

Additionally, the Cloud custom metrics adapter enables the horizontal pod autoscaler to utilize custom metrics, which are often better representatives of system load, as opposed to default metrics like CPU or memory usage, especially for web applications that might not have a clear correlation between resource usage and performance under load.

In contrast, utilizing the average response time from liveness and readiness probes, as suggested in one of the options, may not reflect the actual load on the application or provide actionable data for scaling. This is particularly important as liveness