Open Telemetry Uncategorized, 3 Jan 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2023-01-03 meeting

Description

Instrumentation: Messaging

A

Thank you hi Ted uh good morning.

B

Good morning, good morning, good.

A

Morning, hey um I, I need another. Two minutes uh can I come back yeah.

B

Sure I'm not sure if anyone else is going to be here but.

A

um Yeah, but we can talk, I think you uh had. You know mentioned about the some TC member helping us right. So we can talk about that briefly. Yeah.

B

A

Come back, okay,.

B

See you in a bit.

A

A

Happy New Year.

B

Happy New Year see ya.

A

Yeah yeah I think it looks like uh uh like I, think others may not join um I.

B

Know a lot of people.

C

B

My team are still still off yeah, maybe they're still on vacation yeah.

A

Even for us officially, it's off today, um it's starting tomorrow, yeah.

B

um But yeah what what's on your mind, what would you like to talk about yeah.

A

um Well, nothing much I think.

A

We are making, uh you know, slow progress, but lately I think um you know, our data model is converging. um I, think the there are two pieces uh that are um still to be: cracked. One is like kind of half cracked, but it's the the basically the support for uh nested attributes in a support for nested attributes, so that there seems to be um logically, you know, agreement uh about having support for necessary attributes, at least for log records uh in the spec, but how it will be implemented.

A

I think that there is a lot of confusion. I think that, uh but at least there's an active discussion going on the other one which is, um which is a little hard to crack, is, is the ephemeral attributes in the resource, um and it is for that.

A

um You know I was wondering if, if we can have some TC member, um you know give us some guidance.

B

Yeah so um I don't think tigrants back from holiday yet, but when I last spoke with him, he agreed to to join this working group because there's so much overlap between this group and the logging group um well in general. I would like there to be two TC members as a part of every group going forward. So uh I.

C

B

uh This together in a doc uh that I went over in the spec meeting earlier today, but just to to just send you a link to it. Come.

C

B

Google Docs uh here I'll just post a link in the chat, but this is just a high level, doc kind of going over a lot of the things we've been talking about about how to improve the spec process um like what our our issues have been around. You know: slow feedback because we're spread too thin and um working groups getting very deep into their work without without um yeah a lot of interaction from the TC.

B

uh So it's just kind of reiterates what I think the problems are and what we need to do to solve them um and I care about the semantic conventions, because I think we have a lot of them to do and we're never going to get them done. If we, if every group has to go through like I.

C

B

This group is a special case because we really like rum is, but it is much bigger than just defining like what attributes MySQL should send on its client. You.

C

B

Like I think we we really have to had to blaze a trail in a way. Hopefully other groups don't but.

A

B

Http working group took like a year or something really ridiculous, um so but I before we start any of these new groups, I really want us to finish all the work that's currently on our plate.

B

So we have http browser I would say, is the goal of this group right like there's still IOS and Android, but I think we should reboot the group a little bit um in terms of getting it up onto this process where we can say like okay, let's get TC participation, let's get like a a deadline um for when we're going to uh present, like the various pieces that we have for public review.

B

um That's ideally short, you know like six weeks um and see if we can can hustle well, then getting the community prepared to to process our stuff in six weeks and making sure that it's not going to get held up because the spec approver Community is doesn't have the um like subject matter, expertise or something like that. So I would like to see us be like kind of like a test run for this stuff, but the person, the TC member we need the most I think is tigrid.

B

We should have two, but because tigran's running the logging stuff.

A

Yeah, it could be taken alone, too I. Think at this point um it's a set. You know he's. He is not familiar with.

B

A

The browser background.

B

A

B

Exactly like there's uh I think the reason why some of those um discussions in the logging group are difficult is is all very the use case that we have is like very abstract, abstract to them, correct and and I. Don't think we should spend all this time in the log working group trying to like you know, I I, think getting tigrant over here and then getting some just like real world examples of from the browser of like okay. So we hook into this browser event, and then we want to dump like this data object. You.

A

Know like this is what we're talking about this.

B

Is what we mean by an event? You know these are like real, concrete examples, I think. If we could do that with some TC members, you know, uh then it would stop being abstract and it would become very concrete and I think that that will help like everybody. Every time we've had these kinds of issues in the past.

B

um That has been the thing that gets us past. It is just switching to like actual prototypes and real world examples, so I think if we could do that with tigrin, then we'll at least know what we're shooting for there's some side issues. You know right, like nested attributes and logs uh I, think we clearly need them, but there's a question of like do. We also add this to um other signals.

A

C

B

Yeah like span events or like span attributes like the one I care, the.

C

B

Span events, because if we do this just on the log side, now it's no longer possible yeah to go both directions, which maybe we don't need to. Maybe we just want to say span events are becoming logs and like that's it, you know, but I think that it's like that's.

B

We should like have that discussion as.

C

B

As like the that's the kind of thing where it's like, we need to have the discussion, not as like the rum Sig, but as like the the hotel core spec like do. We want these things to be equivalent actually.

A

That we're not yeah, that's a good topic, um so I know um we like. We want to be able to use any of them interchangeably uh depending on the preference, but um I think one thing I could do is like you know,.

B

A

If, if we were uh to not have that choice, you know which way would we go? You know at least we could have some opinions. uh You know that, may you know help you know avoid that problem like we don't need to solve that problem right um totally.

B

A

But it's not there yet, as of now, it's.

B

Not there yet, and the answer could be no. We only want this to be we actually. What we want to do is deprecate span events officially, as a community deprecate span, events which would be fine, they don't go away right, like we, don't ever break things or take them away, but we could deprecate it and say.

A

Actually, that's uh we could go that route. um Maybe we should create an issue uh to understand that who are the folks that implemented span events today, right and I actually looked around in the in the auto instrumentation. uh You know: repos, Java, art, instrumentation and even python. I think and I saw like only one use case where um you know they were using span. Events in the Java case.

B

A

So it's not very heavily, and even that's a you know, some specific library is behavior. So it's.

B

A

uh Like a super big deal, uh without you know much knowledge further, you know I think it feels like it's not a big deal to remove it.

B

Yeah use it end.

A

C

Auto instrumentation.

A

Is supposed to um give some reference examples right, I think, right that the you know the manual users annual instrumentation would follow.

B

Find there's also the shoe of um the one reason why I believe we don't have nested attributes? Is there are a number of tracing back ends that don't support they.

A

Don't support it that kind of tastes actually I have that.

B

A

Yeah I want to ask that question um um that I think um why? Why is that? An open Telemetry is concern uh because open Telemetry, you know like if you go back to its initial uh days when it started. You know you don't like you want one standard right. You don't want three standards. You want one standard, you know the opens. There was one more right: open, sensors.

B

Open sensors so.

A

So those were kind of you know put behind and uh the community adopted open Telemetry as the new standard, so these Zipkin and uh Jager I think these implementation. These are tools right. They are not standards by themselves, so they should adopt open Telemetry, so they should evolve.

B

A

B

A give and take you know the I think what we we want to find: a balance with open Telemetry of supporting what's already existing. If.

A

B

Be a successful standard. I think this is just a lesson that I've I've learned. If you want to be a successful standard, you have to be willing to standardize the practices people already have so, in other words, open televentory, never would have gotten off the ground if, at the beginning we said here's a new data model, that's like not very compatible with any of the systems.

A

B

Is three years old yeah? So so we we do add, um features to open Telemetry uh around. uh But the question is what what happens if someone can't adopt this feature and as long as they're like optional? That's that's! Okay! So like um exemplars I, don't know right. That's that's a really awesome feature, but no one has that.

A

No one has that right, yeah, no.

B

One has it, but it's, but it's awesome. However, it's not going to break any metric system because it's there yeah, whereas if, if we change our data model for span attributes all.

C

Right that actually breaks breaks a.

B

Bunch of systems.

C

B

So that that's that's, why I mean it's just what what things are we going to get the community coming back to us and saying, like you, can't do this, because this is really going to like create a huge computational overhead for us or just just like screw us up the same thing with metrics right when we're defining like metrics like?

B

What's our metrics look like we went through multiple rounds there, because actually it would be very bad if we came up with a metric standard that, like Prometheus and other platforms like didn't, weren't, very compatible with. So that's. Why that's? Why? uh But I think it's it's safe to say like if the logging Community is totally cool with nested attributes, which I think they are yeah, then we need them and we should add them, and you know, maybe it's just the case that yes span.

B

Events are a one-way, one-way Street and what we're saying in the future is any tracing system that wants. These events needs to ingest the log pipeline as well, um which is like okay. Maybe there are also like events that tracing the system up to throw away because they don't have a trace context.

B

Trace context itself is like burden some overhead that maybe our protocols can, you know, eliminate yada yada, but but I would probably be satisfied with that answer if it meant not breaking breaking people, but.

A

Those are the kinds.

B

Of things where we get stuck right, the ephemeral attributes is the other thing where it's just like. That's not something. This group this group can insist that we need an answer that isn't just staple this information onto everything like like we need like like like this is there's a critical part of open telemetry's data model like open Telemetry, has to be able to model this domain properly. So there has to be a place to put things like Spanish session IDs like this like, like it can't just be an attack on, but it.

A

Turns out, that's like.

B

Really that the open Telemetry Community has to accept that this is a requirement and then decide where we're gonna put it, because it's, unfortunately not it's not as easy as just defining a a session ID resource and moving on. If.

C

It was that easy. They.

B

Would care- but this is why we need TC members in here, because all those conversations just just just kind of run aground um so I, don't know those are the kind of things I'd like to see us get get hacked out once we have at least hegren involved, and we can maybe set a deadline for working.

C

B

With tigran to come up with like a new version of all these proposals, that we have and then go back to the community with them, and hopefully.

C

B

Deadline is short, I'm I think this group is very active, so I actually think we're a good candidate for saying, like we're, gonna hustle on this We're Not Gonna We're.

C

A

B

A

Yeah, okay, that's good to know so! um Maybe next few days uh you know, whenever you get a chance to talk to him if he can join even next week yeah. um You know he said.

B

He was going to uh going forwards, but I just I'm, not sure how long he's on she's on break four seems like he was on an extended leave yeah, but oh I'll ping him to to see when he's coming back nice cool man well,.

C

B

Much for your leadership in this group, it's been really helpful.

C

Thank you talk to you later.