Open Telemetry Uncategorized, 10 Jan 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2023-01-10 meeting

Description

Instrumentation: Messaging

A

Hello, everyone.

B

You know everybody had a good New, Year and holiday break.

C

A

B

Like sleeping in until like 10 and 11., really.

A

Do you know if um Ram is joining.

D

B

He should be uh I guess that was my first day back. We didn't get a chance to sync up. Let me see if he's online.

C

Relative to scrolling.

B

For one of my list here.

B

He's showing us in a meeting that could be this meeting on the downside of team is that's what it says. Oh look, I've got a meeting scheduled I'm busy, you're online or not.

A

Okay, um we can get started.

A

So, um in continuation to what we were discussing uh before we started the Break um I I think we can go over. uh You know this. uh You know this document uh that I had started working on I added one table at the bottom. uh It's not going there where uh I I tried to you know describe what are the different types of objects now, uh whether it's an even span or an event or a span event uh and and then I wrote some comments uh and the purpose is only for discussion.

A

I'll remove this um I'd. uh Maybe it's needed it's useful, but I I don't see at this point, so we can. We can go over. You know this talk. You know this table um is that okay.

B

um I guess this is the effect of that the spreadsheet right so sort of clarifying when, when we send this type of event, we'll also have a span or we'll only send this as a span event. Or is that what you? Where are.

A

You going correct, correct, um uh yes or no, so basically uh you know I I um I had one question: I sent you on slack.

A

No, that is okay, I'll I'll, uh you know repeat that here.

A

um So there was uh a time I think in one of the earlier meetings uh you had mentioned that uh you know the reason we wanted to have a page view or a page impression event. In addition to the you know, if the document load span that exists today, you know was so that you know you know you you want, uh like you know, people redirect uh across the different pages. You know quickly, so we might lose. uh You know that span, so you want to use the first opportunity to a list opportunity to send.

A

You know some event that the user has visited a page right um and in another uh meeting uh a few weeks ago. I think you also mentioned that you know this event that we generate. You know it it doesn't. It can be emitted right away from the instrumentation, but the SDK will still send it along with the you know, rest of the spans whenever the browser close happens uh or a browser. You know navigates away to the next page.

B

But it'll be a little bit, so that's it different thing.

B

So, depending on on how you're running you'll you'll either have a span that you want to represent all this and then you'll have the page view event that will be fired off and and sent um you're then got. You know, other resources that are load and then effectively, you'll you'll end the document load span and that would be sent off right. The the sequence we're talking about correct.

A

E

um I I think satosh is asking if, uh if we're gonna badge the page view event as well as the document load span, um you know. What's the purpose, you know we're not sending it up right, so I believe that's what the misunderstanding means.

A

Yeah I mean it's good to sit in the queue and only sent uh you know, along with the other uh items right.

E

I, don't believe so. Ned can uh correct me. um I.

A

Mean that's the default Behavior. Uh If we use the standard uh like fixed uh time interval of sending uh of clearing the queue from the you know for the purpose of exporting, it only runs every five seconds or so.

B

But there's two different queries, because this event is a log event right. So this came up in the spec meeting this morning, uh kind of clarified some of the um one of the defaults uh and wait man random service, but but effectively, uh and your uh PR I think Riley brought up uh about effectively having one export um so effectively one one batching mechanism um that got shot down. So that's probably not going to go anywhere. So there will be multiple connectors, so you'll.

A

Have that no, that is fine but I uh what I'm trying to get to is like for the events for some type of events? Do we want to skip batch batching like do we want to send it right away? You know over the wire.

B

um Yes, so the the strawberry priority so um I think logs has a priority. Does it not? No, um because yeah Daniel's got a clarification PR out at the moment about the the batch span, um and that can also come up with respect meeting this morning. I'm, currently, writing up clarification around that to try and reword it a little better but yeah.

B

There needs to be some concept of uh priority, which I thought logs had definitely had a priority in there lots of a log level, log level uh that might be what I'm talking about yeah.

A

B

Not it's not exactly the same thing: yeah.

B

It's something we should have and I guess Ted, um where we were talking about the the 200 milliseconds that would come into play. um So you want the higher priority items to be sent quicker uh for the for the live processing, Etc.

F

B

C

F

We figure out how to make our current system coherent before.

C

B

Based on that discussion, yes, foreign.

A

Also suggested that we can add a priority flag and you know in there can be two exporters. You know you know. Each exporter knows what uh to take and what to drop so one will, you know, take the you know, items with you know, with the priority set to one and another will take up items with prior to set to zero yeah.

C

A

One will be doing batching, the other won't be doing batching.

B

Yeah batching really should be done, based on a priority level, at least for a client.

A

Yeah, so uh it doesn't have to be in the uh in the common semantic conventions. You know we can uh I I believe at this point it's only a browser instrumentation requirement, so we could just have you know some parameter uh somewhere uh to indicate the priority. Yeah.

B

Because we're going to get to a point where we need, we need to send the higher priority elements first, because during an unload we have a very fixed limit of 64k to get the stuff off the Box. So if we don't have a priority, we're going to end up dropping potentially the vets that people want to keep versus ones that they don't really need. So um we can clarify that later But to answer the question: yes, we I would expect the page view event to be sent pretty quickly.

B

B

um This sort of comes back down to the batching stuff that I'm that Daniel created for the spans, where the wording needs to clarify it so that um it goes around. But trying to do that, because you've got a couple of comments. There.

A

And at this point we need it only on events right uh not on span.

B

I'm hesitating because I I think it should be there on spent as well, but at this point in time, we've only identified the Revenge.

C

B

Because it really does depend on the app so like I've got internal customers like by default. The Microsoft sdks have everything as a normal priority and we let the users choose which events they want to clarify in the higher priority um and for most customers. Normal priority works fine, but we have a couple of very specific sets of cases where they have events, they're, really critical about which may or may not be a span or may or may not be an event.

B

um So generally, no, but there are use cases where, yes, it is required.

A

A

I think this document is not well updated, so I'll go back to the notes. Here.

A

um So the next question I have is uh like: when is it helpful to have the timing uh events a separate event as compared to an event within the span.

B

um I would have thought the performance, navigation timing is always an event. This comes back. I I really do want to have a mapping to say we're defining this event that looks like this and that can go as a log event or a span event like I I. Don't.

A

No so that we already did right, I. Think at this table um like, for example, here I do have a page navigation timing, yeah- um and you know this is for the for the page navigation. This could be for the you know, the other resources, so the event is defined, but I I uh from an instrumentation perspective.

A

uh You know: are you really thinking of uh giving an option exposing an option to send it either ways? So when, when is it ever like? Who would ever use uh the option to send it separate as a separate event compared to span event.

B

um That would depend on the vendor, so I I would think these are completely different, instrumentations or instrumentation a would send it as an event in transportation. B would send it as a span event.

B

um I wouldn't have think that that there's an option to say I'm just gonna have an instrumentation with a flag that does it both ways, because that's like just extra code to do two different things: we're only ones ever going to get used um and I I think it like. There's a stepping stone like right now, I, you know everyone's using span events, because that's all that exists so I think it's going to be a case of as a stepping stone to get to a normal event.

B

It would be you know the the instrumentation that includes this as a span event. Initially, um until the logs implementation gets fully deployed, and then it would be a log event.

A

um But but would would you folks at Microsoft, um would you be enabling that uh by default.

B

The way we do it today, it would be an event. So if we compare to what we have today, then it would always be a Blog event, not a not a Spam event. What will we end up doing that's yet to be defined, because we don't have log events today, um but yeah I I would lean towards the logarithm.

A

B

Because that could be a baptism set on its own, regardless whether the span got um uh sampled and thrown away. You'd still get the event.

A

Okay, um so it is this item.

A

I think if we end up uh sending as a log event as a standalone event, then that event needs to is, is just the you know: span ID Trace ID sufficient for linking.

B

uh I would think so because that's the whole idea of this and ID Trace ID, in fact they're, probably the same Trace ID. You may want a different span ID to represent it hello. Man would. You depends on your back end and how you're mapping it okay, they should all have the same Trace, ID, okay,.

A

Actually uh is the reason still the same, where you uh expect the span to be completed and available to export sooner than the the navigation event. The timing event.

B

um I I thought we discussed that in the the document load event would actually run up until page load. um So therefore you could have navigation timings. Potentially they tend to come up to Facebook, but you could have um navigation timings beforehand.

B

You could also come after as well, so it really depends. You could have a page load and then go off and keep loading resources so.

A

For the resource, though, uh what span would it uh be linking to? Is it still the same document load.

B

It depends on the resource. um There is no hard and fast rule. It depends on the application. It depends on the page, um which is why I wanted to find these as as events, and then it would depend on the the implementation in terms of what's available so as part of the instrumentation that does the document load, it can pick up and send whatever it's got available.

B

There might be some other instrumentation for react or angular or whatever that sort of says. Oh now, I've gone off and done this I'm going to go. Do this or it starts a timer and says okay now I'm going to go and see if there's any new uh resource timings that I can then send off as part of this um yeah yeah, it's it depends.

B

It's about my daughter, my daughter was was talking to me over the holidays and I kept answering my questions. It depends I'm continuing the trend, it really does depend on the um on what you're doing uh and what you're doing is the application, not us to finding the standard.

A

C

B

uh For example, like user action, we have the user action that that user action could be triggering a the you know: a spa page navigation which is going to trigger a bunch of resource loads, so you'll probably have, depending on the um the framework you're playing with okay I, clicked on this link, I'm going to start a span to say that I click on this link because I'm doing a navigation I go load.

B

The the templates from the servers um so I've got like you know, fetch requests going uh and then I render the page and then I say I'm done so.

B

Depending on that framework, there will be a span and then there'll be a bunch of events that get fired as part of it.

A

Okay, further for the resources um you know, do you want the event uh to be uh linked to the the document load span or or to another child span of it like currently, I think it is, there is another child span, uh I think it's called a resource switch and the resource timing is a span event on that. Do you want to keep it that way, or do you want to just link it to the the parent document load span?.

B

Well, it's again it's going to depend on when that happens, um like I I, don't remember the resource fetch span. What's the case for that.

E

I think yeah good, a good topic was going to bring it up uh either today or tomorrow, I think um so we need to as a as a group, we need to figure out how the linking is going to work and stuff right. So the question that uh sometimes asked uh that's a default. The behavior that we that we have uh in Microsoft um is the document load um in a span is what everything that happens in the page is going to be attached to.

E

um You know if we kind of visualize how this might look for a spa page where that runs for say, 31st or something like that. You're going to have a really weird three single level. Pretty much is it really useful?

E

um Is it going to overwhelm uh the systems uh say the backend processing systems or the ux, or anything like that? um It seems like we're. Trying to you know fit square peg into a round hole. We probably need to. You know, think about it. A little bit better. um I think you know.

E

Some customers internally started using this kind of a concept in our internal sdks, and there are one of the issues like that: they're seeing you know really overwhelming amount of stuff showing up on the UI, and the backend system will also prone the number of children child spans to a thousand or something like that. You know there are use cases where it could be more than thousand. So that's a actually a really good topic. I, don't know if you want to set aside time and discuss about that. You know what is the recommendation from RC.

E

um You know distributed.

D

E

You know for services, it worked beautifully, you know, I, don't think we you know with all of the restrictions and stuff the back-end systems might have works fine, but when you take client things and then try to slap it on top, there may be some issues.

A

Okay, so so you want to the.

B

More directly answer your question, though, before we interview that that topic, um the event would be tagged with whatever the current span is so if the current event happens to be the document load expanded or the document loads fan. If it happens to be a resource fetch span, uh if you've got a resource Venture, it would be that so it would use the context system of the framework to say um when I'm firing a log event I'm going to effectively given the the span that I should be linking it to so.

E

Span also, um should we explore, you know uh linked links uh for for creating these kind of links, uh loose links or whatever. If systems want to really create that uh you know a single level three, you know really big. uh You know single level, three with that thousand plus branches or whatever they could do that um it's should we explore options like that lid stamps.

B

um Yeah because the document load Span needs to end like until the span ends, nothing gets sent. So it's you know the whole based on this name. The document load span would start on. You know pretty much as soon as the SDK is initialized and then, when n, on a page, load or or some render trigger, depending on the framework, um whether you also have an overarching so whether the application also have a overarching.

B

This is my user span um and then the document load span and your links to that user span um would depend on the application. But that's where you start getting into the the child spend stuff that you're talking about Ram, where Okay, um if you've got a user span, um which then has like it's a spa. So it then has the document load. It has user interactions. It has your page flips. Your dialog displays your fetching of resources, you're quite quickly, going to end up with more than 15..

B

um But likewise, if you want to link them all into a tree- and you don't have that user span, then what is the parent span that everything is linked to so yeah? There are some challenges.

E

B

Which we need to talk about for I.

E

Think the the placement gets more pronounced would be if the document the page is passed in the context from the server yeah, um then you know. um That's that's basically depend that you have to make sure that yeah that we want to simply just uh create everything, as you know, everything as a first level child and after that span, or are we just linking that to everything that we create inside of the page.

C

A

um So how what is the current behavior, though uh Tito and Martin? If you know uh the current behavior in the document load instrumentation, does it generate resource which, spans later, if resource is loaded after the document load, is complete.

G

A

Don't think it.

G

Does I think it's uh when the document uh the window load event fires then, but that's when all the spans are created, yeah.

F

G

They're created from from the from like the events from like the.

C

D

G

D

Why they built the next documentation just for later ones,.

A

Oh so there's a separate except.

F

It doesn't link them to anywhere, okay, okay,.

A

Okay, so then um we could uh in in the new scheme, we could uh do it in such a way that all the resources and- and even you know, the Ajax calls uh xhr and fetch are any other Ajax calls. All of them, you know, could be separate uh traces, independent traces and they could all be linked to the you know. The the initial document load span.

A

A

Would that work.

D

I mean we've been doing fine with just the user session ID and Page instance ID. So.

F

But it seems like people to ask for it in the select and stuff.

D

Yeah, that's down to people wanting to already existing servers at tooling to also use the front end which separate question. How good is that yeah.

A

I think one one advantage with using linking is this panel links are part of the standard, so uh they will have a wider support than through sessions where session is a rum, specific concept, um so any you know tooling, that would be you know.

A

Built will be specific to Ram, whereas links would would have more wider support like, for example, a backend uh service could spin off uh jobs right it and and the you know, each job could have a could start at race, but they will all be linked to the original address, so it it. You know to me, you know they they all they these two scenarios. You know they are similar.

E

Yeah um messaging I believe is yeah yeah.

B

um Just a wild thought extending that idea.

B

We could create a session span and then link everything back to that.

D

uh Yeah, it's it's! It's that you have the issue of People, respecting it's sent to be when discernments yeah.

A

Exactly I think.

B

A

The time yeah exactly the session could still, you know, uh be a completely it could be there, but completely independent to all of this uh I think the the only thing we are capturing here is uh the relationship to relationship with the initial page load, uh irrespective of whether it was part of uh you know session one another session, two that that session information could come as a separate concern.

E

Yeah I I think so too yeah.

B

Yeah because I know today, everyone links are based on session on the back end, uh which the teacher was talking about as well. So.

G

A

um That's okay, I'll capture uh both for now and we'll we'll decide. Yeah yeah.

B

I guess my only concern I was just thinking which is really quite um if you've got if you've got a spa and you want a new page will be a new session because they've signed out and signed in or something you probably would also want to change the link, because now it would be you don't necessarily want to, or you may not necessarily want to link it back to the document loads. The initial document load screen, I, don't know it would probably depend.

B

uh Because yeah generally, the teams I work with it's really the session. It's the grouping mechanism for linking everything back together and then trying to repeat a picture.

E

So the the use case of the server that got the original request to serve the the page, the web page, passing down context to the page seems like you know it's it's it's a.

D

Theoretical thing are.

E

There teams- or you know, uh companies that actually use that um uh you know kind of relationship at all.

E

I know we've been talking about this as a use case um and uh The Meta tag way of passing the context down and I I also remember chatting with Santosh about it. You know several months ago, but at the moment do we have use cases like this. um Are there requirements.

B

Yeah, it's called Advanced correlation. um If you go searching for application, insights, Advanced correlation, not very many people, do it.

E

What about um what.

A

Was the question I I.

E

uh Maybe maybe I'm not explaining, but I'll just you know, you know, give like you know uh the you remember the meta tag right. So uh you you request um a page. The server receives a request, the server point of view, uh the server the web server is, is most likely going to create a.

A

Oh yeah, the initial page yeah yeah.

E

A

That doesn't generate a trace, a trace ID from the browser. So how do we solve that problem? Yeah.

E

So they could pass that uh Trace context down to the cursor via the meta ties um uh are the headers. That's in discussion.

A

E

Yeah server timing into the browser, then the browser will have to use that as the context and then um you know propagate further when it makes rejects, calls or stamp all videos. That's the uh basic- and you know uh use case I. Think um correct, is, is this? You know how frequent is this uh use case? You.

A

Know well, 100 right: it is needed uh uh all the time.

D

A

D

Everyone was rum and type game products.

F

They link them together.

F

It's been uh isn't like a parent of the okay.

E

So how do you guys then uh deal with the reason why I asked that is because uh I you know we have capabilities for doing that like when they've mentioned? Not very many customers use it. How do you deal with cast pages and stuff you know? Do you still have um the same context going to tens of thousands of cash instances that you send out or do you change them or you know? How does that work.

D

So we so at the start you mentioned that.

C

D

The server side, the trace IDs in the client side too, and we don't do that we actually skip them separate because in in addition to just the page load, there are so many static resources which can have a trace IDs by, for example, an nginx server. We.

C

D

A plugin that also generates use it for that so yeah we just keep them divided, I, don't know. If Santosh has the documentary really AC accessible, where we talk about the older ROM events, but I did include a few screenshots of our user interface, where it shows how we do the linking.

B

And I'm just trying to find it because the I think the existing document loads, the instrumentation already does this. It goes looking back on the page, yeah.

D

It does, but we just ignore that one yeah.

B

You just don't drop it onto the page. Therefore it doesn't happen. Yeah.

A

Right yeah I think due to your sent it to me some time ago, yeah I like to find it, um but that that's uh you know one problem: uh RAM I think uh before before that I think uh I'm writing the notes. Here I just realized something. If we use sessions.

D

Questions and uh the page instance ID.

A

As uh as a parameter in in all these pants right.

D

Yeah, basically a Christmas level, but resource level.

A

Okay or page instance ID, not just the page, URL, okay, yeah.

D

Just a random in general that ID each time you load of H.

F

D

A

Different: okay, okay, yeah! No! That's good enough, because if you use just a session ID session, ID could change. So I was wondering what yeah already.

A

A

uh Page instance edited we have that.

B

Yeah, we didn't call it page instance: I didn't even work it. We had a yeah.

B

Yeah was it the Service Institute which went flying past I? Think we end up. I wasn't happy with the name, but we're just leveraging the theme. Okay,.

A

Yeah, it is a service instance out here. Okay,.

A

How many different answers.

A

Okay, so yeah I think, let's flip out this, and maybe we can. We can decide uh another time, um and since you talked about uh the correlation with the APM side uh of our the initial page, uh I I think there aren't many options right. There are only two options. You know either that meta tag or the header and I think as part of the spec. You know we could say you know our instrumentation could be looking for both in in some order and whatever the APM sends.

A

You know you would take it and use it right.

D

F

Could you could use cookies, but that's tricky yeah.

D

It's the headers, because you have multiple resources. We also use it for image stacks for script tags for even hsrs to be include linking via that.

A

F

Like an image, you can't link it in any other way.

A

Okay, so for resources, uh what what method Works! uh Only the the server timing, header.

F

Yes, but software doesn't support it yet and last time I checked they weren't too keen and supported me no.

A

That is, okay, I think we don't have many options, so whatever works.

F

In whatever situation, there is literally nothing else you can use so true, okay, okay for the document load, you could yeah. There are a couple of options for for everything else. There is nothing.

F

I have a Deja Vu moment like four months ago or how long have been gone. We had similar discussions.

D

Now you can stop working on the track native spec.

C

A

Yeah this uh this topic needs an elaborate uh write-up, uh so so more folks understand better yeah.

B

I've got some Internal Documentation on this already yeah yeah find what I can grab from that.

A

Okay: okay, uh the next.

A

Yeah I mean I, wrote this last. The fourth one uh I think.

A

You know if, if uh I think that was assuming that you know I think in one of the previous calls uh never think you mentioned that you know you don't need to send this page impression or a page view. We went ahead of time, it's okay, if you send it and the when the browser closes.

A

uh So that led me to you know this question that you know done. In that case, you know: do we need a separate event at all? You know we can just send use the span, but if you are now you know, okay with sending the page view event the moment is generated, then it's good to have. You know two separate uh items. You know one yeah.

B

If I prefer to set this event, because if you're doing client-side redirects, you want very little JavaScript on the page to perform that client so redirect. So if we have to drag in the entire span SDK to do that's just gonna work.

A

Oh, that might still be, you know, tricky um I, I thought you're, the the only Advantage is we don't have to wait for. The uh document load is.

B

A

Because if uh if we were to not even wait for the uh a part of the uh script load, I think, then you will have to serve to different scripts right uh you, you still want some some set of scripts to start running, yeah, uh I I, don't know! If that's uh in a practical is that feasible.

B

um Yeah, so the script that actually does the client-side redirect, if you Chuck that in the body- but you have your SDK in the head, then your SDK will get initialized before the the body gets executed, which is the trick. We use um okay, but some people also put it in the head. So as long as the SDK is loaded before the script, that redirects away is run, you'll still get it and you'll still get all the events.

A

Okay- and that could be an advanced, uh a use case where you.

B

Know we might need a specific cut down bundle for them right.

A

Right: okay: okay, let's skip that uh for now. um Let me get to this View ah anything else uh that stands out to you. Folks, looking at this table, I I have some question on the web vitals too, um but I on, but I think other than that uh you know do you. Anyone has any thoughts comments further on this, so.

B

I'll talk about the question about the resource fiction.

A

Yeah, so the resource uh fetch, uh I I think we kept it pending right, as in uh we'll have to think through and uh get back.

A

But in in either of these two options that I have listed I think it will, uh it will still be a separate Trace independent of the the original uh you know trace for the document load yeah, uh it's just that how it is how they are linked. Is it via the you know, spam links are via the you know, the session ID and the page instance ID.

A

uh You know that you know we can we can decide, but but the current structure of uh the the resource fetch being a child of the document load that can be uh taken out.

A

That way, you know all the resources, whether they they load in the beginning or later they're all modeled the same way in one in one way.

A

Does that uh that makes sense? Does that work.

A

And and even the uh today I think the.

A

There is also a document fetch I I, don't know if that's required, you know we could just have the document load and the document load or whatever we call it. You know there can be only one span uh for the document load uh that will include. You know the the the navigation timing, the page yeah, so this this can be like I I didn't put anything for the uh in the First Column. You know this exists today. uh The second column is the current name.

A

They exist today, but in the in the new model we are, we are taking it out.

D

I would make a case that both document fits and resource fits would be just the HTTP span.

A

D

Because uh so document load would be the one that represents the page life cycle from starting load to upload event at the document. Fetch currently represents the actual volunteered page itself from the server, and it ends the moment. The response ends, which is right before document that uh before way before load event the fires, so it is more closer to http.

A

Okay, so, um but but uh do you want that to be a child of the document load? Are you is? Is it okay? If it is a separate.

D

So, do you want to say, do I say nothing about it, which means that automatic uh parenting.

E

D

No apparent because I would say automatic parenting is better because then you can then the separate there's a separate Logic for if it's in the page load life cycle, or it's maybe like uh during kind of Click span with that parents record.

A

Okay, okay, so let me write that term.

A

Okay, uh but this is uh this- is implicit right like we will. We will still link it uh to the the page load uh you're using this other mechanism. We spoke about.

G

D

Right so I just assumed the other mechanism is like Global for everything, correct yeah. It's implied: okay, okay,.

A

Okay, um that's all, then my questions, oh one web vitals, I, think web vitals, I I just captured uh uh the web vitals event 2, but you know one question I have is uh it has uh like you know the the uh the implementation that you sent to me. uh T2 I think it captured only the name and value but I think the the latest, uh uh the web, vital documentation I think they also support some attribution events.

A

In in I, think that can tell you, like you know what caused the the the value to increase, or you know decrease, which particular Dom Elementor or some specifics about. You know the you know why that that event was emitted. Do you want that to be included? I think might be helpful. uh I'm talking about this.

F

I think so: yeah yeah yeah.

G

Right at least some of it I'm not.

A

Sure yeah yeah I'm not very sure about this, so maybe uh yeah.

A

Okay, yeah I'll I'll make some more changes and, uh and uh one other thing uh uh you know these samples uh nav. uh The two options that I uh put here, uh tigran confirmed in in the last week's uh Sig meeting, that the option one is not possible today did.

B

I yeah exactly um yeah, it's more understandable in terms of option one, um and if we could get the cloud events definition, we would have to support option one, but yet right now today that option one is not supported yeah, but in terms of documentation being able to read it, it's just a lot easier than the awful attribute definition that exists today. Yeah.

A

So so we'll have to go with option two, and what was the update, like did? Anyone have any comments in the spec seek today on on this topic is.

B

It going to go through yeah this one didn't come up.

A

B

A

It was on the agenda, uh I couldn't join, but.

B

Which one, let me see if I can find now, uh of course, my PC uh misbehave this morning, so I joined five minutes late and they're already halfway through the list.

B

B

uh Yeah that already that was the second item listed I joined now. I talked about the third one.

A

Oh no, this is yeah the second item on the agenda yeah because.

A

But, but what what? How does it look like you know? Is it going to go through anytime soon,.

B

um So I don't know I I, I joined after that looks like um Kristen's already signed it off. No one blocked it yet, um but no expect approvers have approved yet either. Sorry.

B

I think this clarifies the need to be the existing definitions. Right is this. The one I read yesterday yep array of map of actually yeah so but yeah. If this goes through, then we've affected the doors open for attributes everywhere, including space. Okay,.

A

There are some people who don't like it too. Yeah of.

B

A

I think John didn't specify why so.

A

B

Yeah I might go back and listen to the uh recording. So for some reason my PC, wouldn't even let me log in until I, had to reboot. It was a whole bunch of delay.

A

Yeah uh I'm thinking uh I'll after some more cleanup of this uh write-up I'll uh I'll join the jsic uh tomorrow and ask Daniel if I can uh submit this as a PR to the JS repo.

A

What what are you focusing? Where should this document uh live.

B

um This one's specific for browser events, it makes sense that it's probably going to be the js3 code. I'd still prefer to see the the event General stuff to be in the Spectrum code.

A

Like uh the documentation, finally I think I I want this to go into the documentation somewhere here, uh so let's say under JavaScript. You know we could have one browser section dedicated to browsers or so all this. Where is this documentation? Do you know? Oh, it's actually.

A

Oh there's a separate, open Telemetry. I o.

A

So this code is not Auto generated, is it I mean this documentation from the.

F

I, don't think so.

F

At least someone was updating it at some point for hotel JS manually, okay,.

A

C

I can I can.

A

Raise a PR against this report too,.

A

Okay, I I can ask in the in the JS check tomorrow what people think.

F

There are not many people who care about browser and says though, but is that in the in the JS GitHub there are people.

B

Yeah, that's really what the sandbox is about, which is what I'm going to get back to this week, is to try and get the sandbox up and going so. We can come up with numbers about whether we can make the existing JS stick code work in a browser or whether we need to.

A

What is this stock in the JS report? Maybe I can.

A

uh You know put it here.

B

um If this may get copied over I do greatly remember several years ago there was some discussion about having the docs Auto updated, so it may be a case of this folder gets copied over to the other one I I. Don't remember. Daniel will know that because Daniel.

A

Yeah, let's ask uh Daniel I'll join the JS SEC tomorrow.

G

So Santa just really quick, you're thinking of putting it in the JS report because uh we're not having much luck getting into the spec.

A

Well, it will go into the spec, also um part of it not entirely uh I. Think some of these uh semantic conventions, uh you know they they will need to be in this spec report, but the the format in which you know the I have put here, I think it's it's useful right. You know the like. You know, I think it's it's clearer here. You know this page. This whole page.

G

A

Spec also is actually okay. Do you want me to try that and if they say no, then we'll think about uh Jess?

A

Where should it go in the spec repo, though I think.

G

Yeah I don't know um yeah I'm, just I'm just asking like why in the jizz um I, also remember that Ted at one point mentioned that it might be worthwhile putting together like an Otep for browser, SDK implementation, maybe no.

A

I I think uh there is no need of an Otep I. Think the folks that care about it are. You know in in this call right, I think uh most others I think the jsig folks will only have a say for the for the pr reviews but I think once I once we make it uh visible to more folks. You know it will get some comments, but I, don't think there'll be a big objection.

G

Yeah I'm just I'm just saying, like the the JS, the JS SDK is just like one implementation of a browser SDK right like so. If somebody wanted to to write their own, that's compatible with uh open Telemetry. Oh that way, yeah so like what would they do like they would need to copy what the JS SDK did.

G

A

So then, then it makes sense to put in the spec repo, but where should it go? I think I am unclear.

A

Maybe in the specification top level I can uh put it here.

F

Asked tomorrow, yeah.

A

B

Yeah, uh because what year was it yesterday in the spec working group, so the semantic veterans work group actually I was in the maintainers meeting. Ted brought up um uh around the semantic conventions, for um you know effectively doing the linking, uh because at the moment uh that, as you found Santosh there's some places where the conventions are defined in other places, where they're linked uh right. He mentioned about uh having a work item to try and unify that so effectively, they're all defined in one place, and then everything is linked.

B

um So wherever that location is is probably where some of this stuff needs to live. I'm. Wondering, though, do we look at like in the specification here we have logs metrics resources, Etc based on tigrants PR that he raised yesterday uh effectively making the events API its own top level thing. If that comes through, then it would make sense to have an event, but.

A

Not all of uh the things in this document are about events, so I think this is more of a domain browser as a domain yeah.

B

Yeah but in the event it's called we're lucky you've got here: you've got logs events, semantic conventions. So really it's a case of. We create a an advanced folder inside the semantics conventions, folder and then you can have browser. So then you get into your domain specific things for for the events, foreign or maybe we just moved the event semantic conventions to to the spec top level spec level, so you've got baggage common compatibility. We just have with events, so it doesn't live in logs, for example,.

B

Because yeah there are going to be domain specific things like the concentrating on browser because that's what we're playing with, but there will be other domain specific. Okay events, okay,.

A

Yeah then, let me uh like actually in I, have put it under events. Only today uh under events under logs I created an event semantic conventions and I I put this there yeah.

B

Let's do it there for now and I'm sure we'll get comments. um Yeah.

A

This browser resource, varying though I, will have to Omit uh in the initial PR, because I think this doesn't fit in.

B

We don't have anywhere for this year, yeah, it did rises in the maintain this meeting as well um in terms of his favorable resource, okay, okay,.

A

C

A

Okay, I think he also said that you know he will have somebody join us from the TC. uh Does.

B

That came up as well yeah to try and make someone uh to allocate to people so, and he did talk about our group, so, okay yeah, he is trying to get that and I think he was actually going to drive that one as well. Oh okay,.

A

Okay, because all of this section I think is, uh we will have to take a call like if that ephemeral resources is not going to happen. Then we'll have to repeat this in all all the yeah.

B

Yeah, it called out there from here's the form of resources PR as blocking us. So, uh although all the maintainers are in the meeting now, no, whether they knew directly or before or not, okay um and he's got a list of things that he wants to try and go through as part of the maintain his meeting, because normally maintain's been pretty short um but yeah I think we treat up like we ran over time to 10 o'clock before Trust gave us the time check um so yeah.

B

It's it's getting more attraction this year, because uh yeah Ted is starting to push it for us. Okay,.

A

But uh assuming uh you know there is uh like we people on this call. You know we're all in agreement uh with with respect to you know the content of this document, except for this varying resources. Then do you think uh you know in another two to three weeks. uh You know we should come up with a plan and start the implementation.

B

Tom I think we can start with prototypes anyway, which I think Martin's already played with, but I think with Tigger and effectively trying to get rid of the apis and move the API. That's complicating things.

B

um I think Java is the only the only one who actually has a release: implementation of the spec. That's now not the same as what the spec is. Yeah.

A

Actually that I don't think we should be worried too much uh nav, because the you know the uh at the end. You know the event. Api will remain right, whether it will be based on log, API or log SDK, yeah I, think uh you know that's uh the internal detail, um but I think that's just the implementation yeah yeah more than that uh I think the data model concerns. uh You know we need their attention on the data model, like the the um the nested attributes, yeah, which.

B

A

I think yeah yeah I think we should have him. You know uh prioritize that over the log API separation.

B

um I think that one came up as well: I I, don't remember: okay,.

A

Okay, yeah I I I'll talk about it in the lock set to tomorrow uh in terms of priority.

B

Yeah children wasn't in yesterday's meetings and he wasn't in the one today. So is he on holidays this week? Oh I know he was here last week, but um the maintainers I think it came up. It said he was out on holiday, so okay, yeah but Jack's been in all those videos. We'll have some context. There.

A

Okay, uh it's 10 o'clock uh here, um oh yeah, we can end here. I'll I'll make the updates and uh send you all a note.

C

A

Thanks everyone thanks so much.