Should factorLatency use c.Threshold as Quantile?

[yflau]

I think we should define an independent variable to represent the Quantile for factorLatency while not use c.Threshold，where c.Threshold should be the latency value?

 		l := hist.LatencyAtQuantile(c.Threshold)
		if l.Nanoseconds()/int64(time.Millisecond) > int64(c.Threshold) {
			isTripped = true
		}

[mholt]

What is your reasoning for that? You might be right, I just want to know why you think so.

[yflau]

As I understand, for hist.LatencyAtQuantile should accept a meaningful value between 0 and 100, and returns the latency value(l.Nanoseconds()/int64(time.Millisecond)) of a quantile, then if the value large than the threshold, should tripped.

For example, if I want it to be tripped when 90% latency > 300 milliseconds, but if I set the Threshold to 300, then the result will be 100% latency > 300milliseconds, which is not what I want.

[mholt]

Your logic sounds correct, yeah. Will let @dannav chime in when he has a chance too, he understands the implications a little better than I do (off-hand, anyway).