The internal duration metrics are not placing entries into buckets correctly. The le label, which is used to specify the bucket, has been set to infinity: le="+Inf".
This isn't very useful b/c most requests are getting put in the infinity bucket.
My initial guess is that using milliseconds is too small of a time unit. Maybe using seconds would work best. The histogram can record float64 values so partial seconds could be published.
My initial guess was wrong. The time.Duration type is actually an int64 representing the duration in nanoseconds. In the code we then divide this by the Millisecond unit. I believe you are supposed to multiply.
Action items:
[ ] Verify that there is a problem
[ ] Fix duration unit conversion
Acceptance criteria:
[ ] (1) The api requests duration metric show multiple buckets with meaningful le values
[ ] (2) The jobs duration metric show multiple buckets with meaningful le values
The internal duration metrics are not placing entries into buckets correctly. The
le
label, which is used to specify the bucket, has been set to infinity:le="+Inf"
.This isn't very useful b/c most requests are getting put in the infinity bucket.
My initial guess is that using milliseconds is too small of a time unit. Maybe using seconds would work best. The histogram can recordMy initial guess was wrong. Thefloat64
values so partial seconds could be published.time.Duration
type is actually anint64
representing the duration in nanoseconds. In the code we then divide this by the Millisecond unit. I believe you are supposed to multiply.Action items:
Acceptance criteria:
le
valuesle
values