Plugin/NeonVM/Monitor metrics we currently have do not count retries, i. e. measure each individual request separately.
Another way to express the same problem is for every scaling session TotalPlugin+TotalNeonVM+TotalMonitor should be equal to total e2e latency.
In order to count the retries also, we must not do TargetRevision: s.TargetRevision.WithTime(now) every time a request is planned. Rather, we update the time once per concrete revision value, and not update it if more request happen with the same revision.
We should also have monitor downscale/upscale metrics separate, as downscale usually needs many more retries.
Plugin/NeonVM/Monitor metrics we currently have do not count retries, i. e. measure each individual request separately.
Another way to express the same problem is for every scaling session
TotalPlugin+TotalNeonVM+TotalMonitorshould be equal to total e2e latency.In order to count the retries also, we must not do
TargetRevision: s.TargetRevision.WithTime(now)every time a request is planned. Rather, we update the time once per concrete revision value, and not update it if more request happen with the same revision.We should also have monitor downscale/upscale metrics separate, as downscale usually needs many more retries.