Open Telemetry Uncategorized, 4 Jan 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2023-01-04 meeting

Description

Open Telemetry Meeting 1's Personal Meeting Room

A

A

A

A

Hey folks, I just put David Ashford to see if he's joining. um If not, we can also skip this meeting.

B

I mean we can start seven I'm, just having some questions to ask.

A

Okay, no worries yeah foreign.

C

Notes we can also.

A

A

So I mean I. I put the first two things. I just want to bring attention to this, rather than discuss these. um So if you're interested in the Prometheus stuff, this is things that are happening um yeah, so the first one is by Russman. uh It is mapping from exponential histograms to Prometheus, uh uh parse, histograms or native histograms, um and the second one is changes to how we handle units.

A

ah Yeah David's here.

C

Hey everyone, sorry I'm late I am.

A

Well, Happy New Year.

A

A

So I was just running through this uh I mean I just wanted to bring attention to these two. um The first is uh mapping from sparse histograms to Native histograms in Prometheus um wrestling. Do you wanna.

D

Yeah sorry uh yeah, maybe it will be nice to get a feedback on it and then I can start with probably writing a specification or, of course, if it's needed yeah as I said.

A

um One thing I do want to call out is that uh we did not do Prometheus exporter, but rather we did Prometheus remote right exporter, because uh the Proto scraping format is very much in flux uh like it's not stable enough, but we are confident about the remote right format.

D

C

Our exponential histogram points only available in the Proto format or.

A

Are they also? No, they are only available in the Proto format and we have to bring the whole Proto format uh into the Prometheus exporter and it's changing constantly. So we were like you know what let's just do demonstrate: okay,.

C

um If someone were to use like the go or the python client Library, are they able to record exponential histogram points there today, or is that also like into.

A

C

Api is also not stable. It.

A

Is available only in go um and only if you use Proto- and none of this is stable, not even the remote right is stable, but remote right is the most stable interface. We could find okay.

C

Yeah, that's fine.

C

And are we also, then I? Guess we're not going to implement it in the receiver until the Proto or text or anything formats are stable, yep.

C

Okay, the other thing we can do depending on. If we're okay, with some churn in the spec is we can specify the current mapping, um but then just not recommend anyone implement it if that makes sense or we can just leave it out of the spec. That's also fine.

A

Yeah I'm not sure about the spec, but I think the idea is once there's consensus on this PR. We will put the mapping into spec and call it unstable or alpha or something yep.

C

Yeah I mean the whole Prometheus conversion. Spec right now is experimental, so.

C

um There's no need to Market, especially if, if we do go through and make it like feature freeze or something, then we'll need to mark this section in particular experimental. If that makes sense,.

C

Yeah, we should definitely add this to the spec.

C

And let me know if you um need any pointers.

A

So for the next one, um I got some internal feedback. We will also make this public on the pr itself, uh but I found it uh useful. uh So basically, first thing is: when do we use ratio uh and when do we use like you know, when there's a unit that says a by B? When do we use that.

A

C

C

So let's see the the a per B comes from.

C

See if I can paste in here, some of this may be us reading into things too much.

C

So the a per B comes from the bottom of the Prometheus naming recommendations.

A

I see, okay, um so again, I don't know uh I I'm kind of okay with your recommendations, which is why I said lgtm yeah, but this came up. It will also be posted to the pr, uh but this is good to know. I will go back to the Prometheus team to see what we can do here. Yep.

C

Well, and of course, we just don't want the slash in the unit, because that gets added to the metric name yep and then what was? Let me go back to the notes, converting one to ratio.

A

That is also uh here. The base unit is ratio, disk usage ratio, um disk usage, I see yep um Let me. Let me go back to the Prometheus team um and get back to you on that foreign.

C

I'm also, okay, with something that's more conservative.

C

To your last section, yep totally fine with something like that.

C

C

And then the last item is preserving Prometheus, unknown type.

B

I have one questions for you. I have one questions regarding the exponential promoters. Is that okay? Oh so it's will be different with like for me test V1 vomitis V2, where it has different format and and also would be the same when you know when transferring from.

B

uh Histogram permitters to permittest receiver as well or like because we are transferring from exponential to histogram and then underhand. On the other hand, we would lose information when transferring back from Instagram to exponential. You know it's upon us always not Guaranty, to have.

B

um If that's is that a question, that's a good question.

A

um So, just to kind of reiterate, like exponential histograms, are being worked on in Prometheus right now: they're they're, completely unstable, um but in the future, when we do get full support in open Telemetry and in Prometheus as well, you should be able to go from Prometheus histogram to open Telemetry histogram to Prometheus histogram without loss.

A

But today we are kind of working with the constraints that the Prometheus project has, which is everything is experimental. Everything is being worked on and we are like to kind of have something. Let's start with the remote right, I fully expect us to also add the Proto scraping once more than one language supports it today. Only the Golan client supports it. Gotcha.

B

Oh I didn't know that the exponent show histogram from Prometheus is unstable. So thanks for that, thanks for the clarification.

A

Oh, no, it was just merged in November, but it that's the merging point. We still have to get feedback from users and all of that before we can call it stable.

B

And would work, uh Prometheus has different kind of schema as well, so just want to clarify, with based on the Prometheus histogram schema, 2 or schema one. So for a special explanation. Histogram.

A

What do you mean schema here? Is it the scale.

B

Oh, no, it's not the scare, but it's just like, because when Matthias has oh look, I just one second,.

B

Oh I can share the screen too.

B

Oh I, don't know.

B

That one uh yeah one second.

B

In the zoom chapter there would be like two.

B

Schema with promoters with the matriculation we want it will like uh so just wanted to ask if, like what.

B

A

uh I I think Matrix version, one with Matrix version 2 are from are like influx or Telegraph related. Okay,.

B

A

Not not Prometheus related, okay,.

B

Thanks for Crime meditation yeah, then I I don't have any questions left foreign.

B

Last one, uh that's true.

C

So sorry, I haven't gotten a chance to review this one. Yet I will try and actually it's not that long. Yes,.

B

Yeah, it's just uh questions like what is the crossing point between using the better flag and using the uh using the.

B

Attributes, though, let me.

C

So I'm having some trouble understanding you what.

B

Was that like what is the crossing point between using the data flag and attributes like if performance issues or if you know like if performance is uh one of the criteria for using the telephone fly, but not using attributes? Because I saw that we informative side? We use like no record value when.

D

B

With like some encounts but just yeah, should we use that with when uh converting from.

B

And type 2 when showing is a type type instead of couch yeah.

C

I would personally prefer something closer to a data point flag.

B

I mean I'm fine with this. What's up, oh, no go ahead! Sorry.

C

It is weird that it's per data point, though, rather than per metric since type, is at the metric level. I wonder if we need a flag concept there.

C

Yeah I'm not thrilled with either the attribute or the data point flag. um So maybe we need to look for alternatives.

B

We don't have any attributes for metric level. We just have like teleport, oh the whole resource Matrix levels. That's why that's a painful being at a data point level. So.

C

I I will take a look and leave my thoughts. Okay, yeah.

B

C

Need to think about it, some more! Thank you. No.

B

C

Okay, is there anything else anyone would like to discuss.

C

Otherwise, everyone can have 40 minutes back.

C

Thanks guys, Happy New Year.

B

Happy New Years thanks Nashville for yeah insight, yeah thanks a lot of time too yeah wrestling see you guys.