The current load-watcher Prometheus pkg was using the metric of instance:node_cpu:ratio to calculate the node utilization
However, when this value is still below 60%, I found another metric instance:node_cpu_utilisation:rate1m was very large and was around 90%. Apparently, the Prometheus metric had some smoothing for the metric, and the one we used may already have a smoothing over a large time window, which might be larger than 1m. Let's guess for 5m.
We are not sure which Prometheus metric is consistent with the metric obtained directly from the metric server, so there needs more testing.
The current load-watcher Prometheus pkg was using the metric of
instance:node_cpu:ratioto calculate the node utilizationHowever, when this value is still below 60%, I found another metric
instance:node_cpu_utilisation:rate1mwas very large and was around 90%. Apparently, the Prometheus metric had some smoothing for the metric, and the one we used may already have a smoothing over a large time window, which might be larger than 1m. Let's guess for 5m.We are not sure which Prometheus metric is consistent with the metric obtained directly from the metric server, so there needs more testing.