Prometheus & Observability documentation


Richard Hartmann
 

Dear all,

during our call I mentioned that Prometheus will be overhauling our
complete documentation, see
https://docs.google.com/document/d/1yuaPKLDvhJNXMF1ubsOOm5kE2_6dvCxHowBQIDs0KdU/edit
for details.

Also, we are working on making 101 content. If you have someone who
can try https://archive.fosdem.org/2019/schedule/event/on_observability_2019/
and give honest feedback, that would be great.


Best,
Richard


Amin Amos
 

Hi Richard,

Regarding Observability 101 content: I noticed the recording page has a feedback link. Should I send feedback there, or post it here?

Thank you,
Amin


Richard Hartmann
 

Post it here, please. I do get FOSDEM feedback email too, but that's
invisible to this list.

On Tue, Jul 21, 2020 at 6:58 PM <a@amos.codes> wrote:

Hi Richard,

Regarding Observability 101 content: I noticed the recording page has a feedback link. Should I send feedback there, or post it here?

Thank you,
Amin


Arthur Silva Sens
 

Hello Richard and everyone,

Maybe some of you recognize me from GitHub, some may not, but for context to everyone.. I'm a recently graduated student with not much experience around the whole observability scope, currently working on improving KubeVirt's observability(CNCF sandbox project). 

I've just watched the observability 101 and I must say that it did give me some really good directions on how to proceed with my project. 
I can recognize some old mistakes that I did with old monitoring projects, often making cargo culting and building metrics datalakes instead of good monitoring solutions. Probably it wouldn't have happened if I watched your presentation back then.

One thing that came to my mind when you talked about Monitoring vs Datalake, I guess that Machine Learning use-cases are starting to grow on the Observability field, so metrics' datalakes may not be garbage anymore. Of course that Monitoring and Machine Learning are two different things, but I think that both have some space at the Observability scope. WDYT?

Another thing that was tremendously helpful for me in your presentation was the clear definition of what I'm supposed to accomplish with Metrics, Logs, and Traces. When implementing new metrics, I often try to solve problems that are better suited to logging/tracing. 

One topic that is still unclear to me is:  When should I need tracing? You say that tracing is expensive and should be treated with care, so how do I know if tracing is worth for a particular case?

Overall, I think that it was a wonderful presentation and I would recommend to anyone who is getting started with observability, like myself. And again, this is the opinion of someone not much experienced with all of this, maybe someone more experienced could give some feedback if what you are saying remains correct nowadays.

thanks,
Arthur

Em ter., 21 de jul. de 2020 às 14:04, Richard Hartmann <richih@...> escreveu:

Post it here, please. I do get FOSDEM feedback email too, but that's
invisible to this list.

On Tue, Jul 21, 2020 at 6:58 PM <a@...> wrote:
>
> Hi Richard,
>
> Regarding Observability 101 content: I noticed the recording page has a feedback link. Should I send feedback there, or post it here?
>
> Thank you,
> Amin
>




Bartłomiej Płotka
 

Thanks Arthur - those questions are an amazing indication for us what sections of our potential SIG 101 doc should be extended (: 

> One topic that is still unclear to me is:  When should I need tracing? You say that tracing is expensive and should be treated with care, so how do I know if tracing is worth for a particular case?

TL;DR the answer is simple: When other signals do not give the answer you are looking for (: 
For example "why my request was slow? What part (e.g which service call) of the request execution was the bottleneck?" I find tracing definition by Peter Bourgon quite useful:

> (...) tracing then, is that it deals with information that is request-scoped. Any bit of data or metadata that can be bound to lifecycle of a single transactional object in the system. As examples: the duration of an outbound RPC to a remote service; the text of an actual SQL query sent to a database; or the correlation ID of an inbound HTTP request.

Kind Regards,
Bartek Plotka

On Wed, 22 Jul 2020 at 01:12, Arthur Silva Sens <arthursens2005@...> wrote:
Hello Richard and everyone,

Maybe some of you recognize me from GitHub, some may not, but for context to everyone.. I'm a recently graduated student with not much experience around the whole observability scope, currently working on improving KubeVirt's observability(CNCF sandbox project). 

I've just watched the observability 101 and I must say that it did give me some really good directions on how to proceed with my project. 
I can recognize some old mistakes that I did with old monitoring projects, often making cargo culting and building metrics datalakes instead of good monitoring solutions. Probably it wouldn't have happened if I watched your presentation back then.

One thing that came to my mind when you talked about Monitoring vs Datalake, I guess that Machine Learning use-cases are starting to grow on the Observability field, so metrics' datalakes may not be garbage anymore. Of course that Monitoring and Machine Learning are two different things, but I think that both have some space at the Observability scope. WDYT?

Another thing that was tremendously helpful for me in your presentation was the clear definition of what I'm supposed to accomplish with Metrics, Logs, and Traces. When implementing new metrics, I often try to solve problems that are better suited to logging/tracing. 

One topic that is still unclear to me is:  When should I need tracing? You say that tracing is expensive and should be treated with care, so how do I know if tracing is worth for a particular case?

Overall, I think that it was a wonderful presentation and I would recommend to anyone who is getting started with observability, like myself. And again, this is the opinion of someone not much experienced with all of this, maybe someone more experienced could give some feedback if what you are saying remains correct nowadays.

thanks,
Arthur

Em ter., 21 de jul. de 2020 às 14:04, Richard Hartmann <richih@...> escreveu:
Post it here, please. I do get FOSDEM feedback email too, but that's
invisible to this list.

On Tue, Jul 21, 2020 at 6:58 PM <a@...> wrote:
>
> Hi Richard,
>
> Regarding Observability 101 content: I noticed the recording page has a feedback link. Should I send feedback there, or post it here?
>
> Thank you,
> Amin
>




RichiH Hartmann
 

Thanks for this feedback!

On Wed, Jul 22, 2020 at 2:12 AM Arthur Silva Sens
<arthursens2005@gmail.com> wrote:

One thing that came to my mind when you talked about Monitoring vs Datalake, I guess that Machine Learning use-cases are starting to grow on the Observability field, so metrics' datalakes may not be garbage anymore. Of course that Monitoring and Machine Learning are two different things, but I think that both have some space at the Observability scope. WDYT?
I agree that ML needs more nuance. I have better mental pictures to
transport my caveats now and those would hopefully enable more
deliberate tradeoff considerations.


One topic that is still unclear to me is: When should I need tracing? You say that tracing is expensive and should be treated with care, so how do I know if tracing is worth for a particular case?
Fair point. I should have talked about trade-offs more.


Overall, I think that it was a wonderful presentation and I would recommend to anyone who is getting started with observability, like myself. And again, this is the opinion of someone not much experienced with all of this, maybe someone more experienced could give some feedback if what you are saying remains correct nowadays.
Thanks; and I think the ones with experience are good at nitpicking,
but, by definition, can't _truly_ say if it helps newcomers. The
newcomers need to do that.


Best,
Richard