There is an alternative solution - to use the increase() function from
MetricsQL - it doesn't extrapolate results and it takes into account the
previous value before the window in square brackets. So it returns exact
expected values. See more details at
https://victoriametrics.github.io/MetricsQL.html

On Fri, Nov 20, 2020 at 2:49 PM [email protected] <[email protected]>
wrote:

> On Friday, 20 November 2020 at 02:29:12 UTC [email protected] wrote:
>
>> *Query*:
>>  1. normal query: error_counter_something{job=“monitor”, device=“dev0”,
>> serial=“xxxxxxxx”}
>>  2. delta query: delta(error_counter_something{job=“monitor”,
>> device=“dev0”, serial=“xxxxxxxx”}[$__interval] > 0)
>>
>> *Time Range*: 2020-11-19 16:16:00 ~ 2020-11-19 16:20:00 with 15sec
>> interval
>>
>> *result*
>>
>> 16:16:15~30 raise 2 errors on device and move that error counter value
>> from 7616 to 7618,
>> but the delta query shows result of 3
>>
>> time                             ,            delta  ,
>> normal
>> 2020-11-19 16:16:00,                       ,                     7616
>> 2020-11-19 16:16:15,                       ,                     7616
>> 2020-11-19 16:16:30,                     3,                     7618
>> 2020-11-19 16:16:45,                       ,                     7618
>> 2020-11-19 16:17:00,                       ,                     7618
>> (keep these value until end of query time range)
>>
>>
> See https://prometheus.io/docs/prometheus/latest/querying/functions/#delta
> *"delta(v range-vector) calculates the difference between the first and
> last value of each time series element in a range vector v, returning an
> instant vector with the given deltas and equivalent labels. The delta is
> extrapolated to cover the full time range as specified in the range vector
> selector, so that it is possible to get a non-integer result even if the
> sample values are all integers."*
>
> You haven't said what $__interval expands to in your query.  It must be at
> least 30 seconds, because otherwise you wouldn't have two values in your
> range vector.
>
> So let's see what happens with 30 seconds.  The window contains two values:
>
> [...X........X...]
>    7616    7618
>     <--15s-->
>
> The difference between these is 2, and the time interval between them is
> 15 seconds.  However this increase is then extrapolated to cover the whole
> window period of 30 seconds, so the value returned by delta() would be 4.
>
> What about if $__interval was 45 seconds?  Then you'd have three values,
> the difference between the first and last is 2, the time difference is 30
> seconds extrapolated to 45 seconds, so the result would be 2 x (45/30) = 3.
>
> If you want the actual difference between the metric now and the metric
> some time ago, you can do :
>
> something - something offset 15s
>
> However, both that expression and delta() will give you nonsense values if
> a counter resets, because it will jump back down towards zero and give you
> a large negative value.
>
> Better:
>
> (something - something offset 15s) >= 0
>
> but it won't handle counter resets as well as rate() or increase() can.
>
> --
> You received this message because you are subscribed to the Google Groups
> "Prometheus Users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/prometheus-users/f75b2007-96ed-4c66-b719-602934827cd3n%40googlegroups.com
> <https://groups.google.com/d/msgid/prometheus-users/f75b2007-96ed-4c66-b719-602934827cd3n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>


-- 
Best Regards,

Aliaksandr Valialkin, CTO VictoriaMetrics

-- 
You received this message because you are subscribed to the Google Groups 
"Prometheus Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/prometheus-users/CAPbKnmAMMFA3gsTAbqVC5uz%3Dp-kymL3QT31vb9G3vOhemjWCfw%40mail.gmail.com.

Reply via email to