Response time in percentage based on a gauge metric

27 views
Skip to first unread message

Sakthi Raam

unread,
Sep 1, 2021, 3:48:02 AM9/1/21
to Prometheus Users
Good day All!

I would like to seek some advice on how to calculate response time (latency SLI) based on a gauge metric. I don’t have a histogram bucket.

Kafka_network_requestmetrics_totaltimems{request=“”,quantile=“0.50”}

Above is the time series of interest which records the response time( in milliseconds) in quantile.

I basically want to get the percentage of request that got processed within x seconds and also want to find the percentage of request that got processed above x seconds for a given quantile (99%- This is available as part of the metric)

Looking forward for some help

Thank You
Regards
Sakthi
--

Best regards

R.K.Sakthi Raam

*: rksakt...@gmail.com

': +65-9185-0354

Linkedin

Sakthi Raam

unread,
Sep 1, 2021, 8:34:01 PM9/1/21
to Prometheus Users
Good day All!

I would like to seek some advice on how to calculate response time (latency SLI) based on a gauge metric. I don’t have a histogram bucket.

Kafka_network_requestmetrics_totaltimems{request=“”,quantile=“0.50”}

Above is the time series of interest which records the response time( in milliseconds) in quantile.

I basically trying to get the percentage of request that got processed within x seconds and also want to find the percentage of request that got processed above x seconds for a given quantile (99%- This is available as part of the metric)

Regards 
Sakthi

Bjoern Rabenstein

unread,
Sep 2, 2021, 12:18:58 PM9/2/21
to Sakthi Raam, Prometheus Users
On 01.09.21 15:47, Sakthi Raam wrote:
>
> I would like to seek some advice on how to calculate response time (latency
> SLI) based on a gauge metric. I don’t have a histogram bucket.
>
> Kafka_network_requestmetrics_totaltimems{request=“”,quantile=“0.50”}
>
> Above is the time series of interest which records the response time( in
> milliseconds) in quantile.

This looks more like a pre-calculated quantile, which is part of a
summary (not a gauge). Although you could argue, the quantile itself
is a gauge. So perhaps that's what you mean. Just saying to avoid
misunderstandings.

> I basically want to get the percentage of request that got processed within
> x seconds

To answer that question, you need a histogram (with x seconds as a
bucket boundary).

> and also want to find the percentage of request that got processed
> above x seconds for a given quantile (99%- This is available as part
> of the metric)

This question sounds like you have mixed up two questions into
ane. The 99th percentily (AKA quantile="0.99") is the duration after
which 99% of requests finished processing, and 1% of requests took
longer than this duration.

--
Björn Rabenstein
[PGP-ID] 0x851C3DA17D748D03
[email] bjo...@rabenste.in
Reply all
Reply to author
Forward
0 new messages