telegraf as all in one exporter

[Yosef Yudilevich]

as i saw ,telegraf client from influx may open metrics in prometheus format. and it has everything in one plase.
i meen no need for 10 different exporters.
why not to use it for everything?

[Yosef Yudilevich]

ah i saw that its antipattern
but its much easyer to manage from ops point of view
and also telegraf has some stuff that exporters doesnt have
like
windows support and procstat per process io, mem,cpu metrics
and say reacher apache metrics

so no coice really here

[Julius Volz]

On Sat, Aug 13, 2016 at 6:11 AM, Yosef Yudilevich <yosef.yu...@gmail.com> wrote:

ah i saw that its antipattern
but its much easyer to manage from ops point of view

It might seem like that at first, but in reality it creates a bunch of operational problems: http://www.robustperception.io/one-agent-to-rule-them-all/

You also lose most of the benefits of Prometheus's automatic up-ness health monitoring and the service discovery metadata that gets attached to time series when you scrape per-process exporters or service endpoints.

 

and also telegraf has some stuff that exporters doesnt have
like
windows support and procstat per process io, mem,cpu metrics
and say reacher apache metrics

Yeah, the exporter is mostly a Unix exporter right now. For Windows, you'd want something different.

We deliberately don't exporter per-process resource metrics, as that is also the job for other systems (it's not really a host metric, and it can grow very large and hard to track when there are many short-lived processes).

If you want service-specific metrics like Apache metrics, again, an Apache-specific exporter (or probably a log-to-metrics converter like https://github.com/google/mtail) are the best solutions.

 

so no coice really here

--
You received this message because you are subscribed to the Google Groups "Prometheus Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to prometheus-developers+unsub...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Yosef Yudilevich]

On Saturday, August 13, 2016 at 3:51:58 PM UTC+3, Julius Volz wrote:

On Sat, Aug 13, 2016 at 6:11 AM, Yosef Yudilevich <yosef.yu...@gmail.com> wrote:

ah i saw that its antipattern
but its much easyer to manage from ops point of view

It might seem like that at first, but in reality it creates a bunch of operational problems: http://www.robustperception.io/one-agent-to-rule-them-all/

why change config counts as a problem , and copy exec together with config, then create startup script for that, and then update it each time, doesnt count as problem?

(in this post)

 

You also lose most of the benefits of Prometheus's automatic up-ness health monitoring and the service discovery metadata that gets attached to time series when you scrape per-process exporters or service endpoints.

 

and also telegraf has some stuff that exporters doesnt have
like
windows support and procstat per process io, mem,cpu metrics
and say reacher apache metrics

Yeah, the exporter is mostly a Unix exporter right now. For Windows, you'd want something different.

We deliberately don't exporter per-process resource metrics, as that is also the job for other systems (it's not really a host metric, and it can grow very large and hard to track when there are many short-lived processes).

If you want service-specific metrics like Apache metrics, again, an Apache-specific exporter (or probably a log-to-metrics converter like https://github.com/google/mtail) are the best solutions.

Developer asked me to monitor Nodejs process that has two threads, for each thread (he need IO of each thread together with memory and cpu, cause he sees it in htop, so how come i cant measure it???)

Also on the same node he needs info about Wowza process and who know what else will come tonorrow

so i can go and write exporter each time!

so i just have no option but to install telegraf and make it export what i need :)

thanks