Open Telemetry Uncategorized, 13 Sep 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2022-09-13 meeting

Description

Instrumentation: Messaging

A

Hi teacher morning.

A

Hey evan santosh.

B

Good morning, hello, everyone.

A

So I I decided to join, I could find the time so, but I don't have anything on the agenda. I did not have the capacity to do anything.

C

uh No worries I think uh last uh last week in this call ram um said he will join and uh discuss on the ram events. uh Nev. Have you folks uh discussed it internally.

B

um Yeah he pinged me would have been on thursday friday last week and he was working on it then. So we found some time to do some stuff.

D

Yeah good morning.

D

Can you guys see here.

C

Yep we can hear.

D

All right, do you want to start with the schema discussions, or do we have other topics? I just brought up the.

C

D

Think we can, we can talk about.

C

The schema discussion, okay.

D

Sounds good, let me share my screen. I um I will go through um the the the skills that people have added there and the old um excel spreadsheet, where we capture attributes- and um you know a few other stuff. Also there have been links. People have added and things I'm just trying to organize that and things now we can start from there. I have a bunch of questions we can go through.

D

um Give me one sec.

D

Let me know when you see my screen.

C

D

Okay: okay, there you go yeah sorry used to uh teams, not zoom that much.

D

You should see now. Yeah.

C

We can see now: okay,.

D

Cool all right, um first of all, thanks for folks that have added their schemas here and stuff, and I removed an organized movement.

D

So I did reach out to a couple of folks over slack to stefan and quinn quinn responded back and he said you look into it. I think it was yesterday or something you responded back. He said, look into it and add the details so far, nothing as far as I can tell stefan, I don't believe- has responded. Yet um we can get this thing started with. So here's here's what I'm thinking we would do. This is a place where we can, where people can dump um their events and uh and stuff.

D

Then I plan on um doing a comparison excel spreadsheet, similar to what um you know. One of the guys I've started before so go down to rom events added a spreadsheet.

D

So, essentially, one in a sheet per uh even schema that you want to. Finally uh do we'll take an example. So, right now it's pretty sparse and stuff and I'm going to be confused. You know we. This is what we need to go through as a team. We can go through each event and start filling up these things. Basically, what I'm trying to do is throw in all the properties or fields for uh every vendor's streamer and try to compare them, and then you know get them all in the same lines.

D

For example timestamp, it's a pretty.

D

Simple thing, but you know you get the ibm uh url for cisco dynamics, compares to uri and microsoft, and so on and so forth. So I you know, timings is probably better because you know martin had already worked on defining some of these things for autumn, essentially, at the end, we would want to populate this uh part of that. We want to agree upon names and the purpose of each of the fields and then see if those fields make sense or not, and thanks that's that's really exercise.

D

I think um we can pick one of these events and start working through them today, uh as we get used to this motions and stuff, and then you know reach out to folks to get their schema information here or basically tell them. This is how it's evolving too, make sure that this will fit your needs. That's basically, you know the.

D

If they are not really, uh you know able to contribute to their schema, then we can tell that this is what the hotel room events are going to look like make sure that this will work for you. That will be a forcing function for them to come. In and try and influence this also make sense, yeah. Okay, um I can go through. You know, as I'm going uh was going through these um stuff that people have already added.

D

I noticed a few and I have a bunch of questions we can go through them. I think um so. We have folks from in this meeting today, uh cisco uh we have francisco. um Do we have you know? Obviously you know microsoft's here um splunk do we have uh you know I sorry, okay and martin from splunk.

A

And you're, like you, did teach me.

D

Literally yeah, okay, um I think events are here. They just don't show up um so.

B

It seems like someone deleted all of the open telemetry current implementation values still.

D

uh I never apparently did that uh he wanted to make sure that we, you know uh understood. These things exist in experimental implementations already.

B

Yeah but that's what splunk currently uses.

D

Oh, I see okay, gotcha, okay, so should we uh probably drop this then if splunk is already using it? That represents um that that use case, I think we can probably drop these things.

D

Okay um and uh for new relic I'll I'll, uh you know add links and stuff here, so that'll be good. Let me go through my questions and then probably um um neural page view looks uh yeah. So the first question that I have is uh page view looks more like timing, you know, uh martin, can you talk to it? uh You know, for example, I'll show you. You know the page views that we have already captured for um come on there. He goes.

D

The drop down just keeps messing up. So if you take a page view here, it's uh for us, it's an impression every every impression we want to record certain pins. You know the timing. Information usually doesn't show up there, it's usually about which url. Where is this coming from and then a few other things that we have for internal purposes.

D

Cisco's also seemed pretty similar to that, uh but new relics doesn't match at all. So martin, you know I'll hand it to you. Can you talk to that number.

A

Yeah, so I think um so I don't know like when exactly you, you capture your page view, but we do it, we do it when the window load, event, fires, and so we we capture, basically, like all the timing information we have until that point on that event, so it's essentially for us, like page view, is essentially like synonymous to the span that auto has for for document.

C

Load, but I think those are uh advanced properties. What are the fundamental properties that you capture, like you, know the the url? What time the page was.

A

Right so I think I didn't. I can't capture all of those, so I think you probably um so like, for, I think I probably captured all of those, and there are probably a lot of additional attributes that you probably don't want to have on page view, which is fine.

D

Okay, I can't see a url here, you know maybe I'm unlisted or something oh.

A

You know what I probably put it on under global.

D

That's what global is? I have another question about. What is a global event uh also, uh I think, uh generally right.

A

Yeah sorry, yeah general, so that's not it's not an event in itself. It's just the attributes that apply to all of the events. Oh.

D

A

D

Block yeah same for splunk, okay, so splunk would be so. Let me just put a comment here: I'll go, try and uh so far every event we talk about. We need to assume that these things exist for every event that you fire, plus even specific right. So for page view, edition yeah.

A

D

A

So quick question on the page view: are you? um Are you thinking that pageview is uh is essentially just like a marker that, like this page, was loaded and it doesn't have any information about about any timing.

D

Correct those could be additional properties, I think so the purpose of page view. Traditionally, I think google analytics also has a similar concept. It's all impressions, so you know how many visitors am I getting for this particular page at this point you know every page that I have you know that's a fundamental measure that every site owner wants. So it's almost bi type stuff. It's just impressions, that's the that's the basic thing that, as far as I know, I'd like to hear others thoughts.

D

Also, what I've seen is that people add a lot of properties to that impression like in the timing, information and things what we did in microsoft. Is we split them? We said there is an impression count that event will be fired, asap um in in the in the life cycle of the page.

D

B

D

Any event, but this one so that's the most critical event for us. You know we don't want to lose count of visitors and things and then any additional properties we can capture and then the things that you are talking about here in terms of um all the timing and stuff, we have another event that happens, I believe at the end, and that can correct me if I'm wrong after the document is loaded, uh will fire all the time you know when did the network connection start?

D

When did the last fight come through and those people stop? So those are two separate events. uh One is pageview another one is um where's microsoft. uh Here.

B

D

B

Free performance.

D

Performance has all of those uh timing relevant stuff, so those are two separate events.

C

Sorry, what what is the difference between page view, timing and page view performance.

D

uh Page view, timing is not our event page view. Timing will be usually wrong.

C

D

Just microsoft is page view and paging performance. The difference is this. Guy is impression. It's reasonably lightweight uh fired, asap, uh usually for counting the number of visitors to the site. That's fundamental measure and this one is for the performance within step. Everything from how is this page performing.

C

And uh martin fyi, I think, last week, in this meeting, um we also discussed uh in addition to the current spans, the document load spans we send, we could, potentially um you know, send an event for the purpose of you know this impression.

D

uh Experiment or a separate event separate.

C

D

A

Yeah, so so, just uh just one more point to clarify on that ram, uh the so the page view page event will be sent as soon as the sdk initializes correct.

A

That makes sense to me yeah, okay,.

D

So sounds good, so then I I think you know uh you know. We can then discuss about some of these things. You know which ones are critical to have as part of that pageview event of these properties. Some of these things might be useful to have as product page view. We can put them as uh additional attributes. I think to that particular event or move them to this subsequent event. I'm open for that. I think that's the discussion that we need to have you know I don't want to say you know.

A

Our events are.

D

Lightweight we want to keep them like that. You know we basically say um um first content for paint, um you know, probably it's it's it's reasonable to capture, so the bar should probably be this. Will you know obviously duration? If you want to calculate duration, you cannot fire the event activity right beyond so this probably does not belong to page view, but other you know same with network deviation. First paint is probably okay to capture as part of page view, I'm open for that.

D

I'd like to hear thoughts about you know, which ones should we do here or do we move all of these timing? Events wholesale to the second event.

B

Yeah there is some timing related stuff like we can fire the pageview, then before we've actually finished painting the page, so it may not be there. um The other consideration is what do we want to do for spar applications so for mobile devices? uh For for us, we actually just have a paid another pageview event, um because we hope the location change event.

B

So you you'll get a page view for the very first page load and then for any virtual pages. We also fire a page view event at that point. um So for some for the for the secondary virtual pages, if you're playing with react or something a lot of these timings are meaningless. uh Really all you've got is a duration.

D

That's that's. That's a good point. Yeah.

C

Also, uh those two first contentful paint and first paint- I I could be wrong, but are they aren't they part of web vitals.

C

Or are they do they come from the navigate performance timing, apis.

A

So they're not part of web web vitals, but they they have their own api.

C

I see, but are they part of the performance navigation timing, api.

A

No um there's a separate api to get that.

A

um There's like a day it's through the the performance observer and like the type of the event is paint, and then it gives you like. You know the one that it has so okay. So it's yeah.

D

Is is, is this something that uh belongs in pageview dns, lookup, english, and it probably is a thing that will be available at that time. Right.

B

That's the navigation, but that one will be there. I would actually.

D

A

Sorry, I was gonna say I would actually all of these things like look like timing to me so like. I would actually argue that to keep the impression just without any timing, information, yeah and and like all of this. All of these are most of these actually come from the navigation timing and the the first paint and first content contentful paint.

A

um They could be like a timing, separate timing event, that's what we do at neural, so.

D

Okay sounds good, so if that's the case, then uh we'll probably okay, let me just capture this in the comments. Real quick.

A

We we initially implemented it as as attributes on the page view event, but then we uh realized that sometimes you know like we, we fire the pageview event when window load happens and- and uh you know, like the the paint the paint may or may not happen before it could happen like like much earlier before or which could happen afterwards. So we actually send it now as a separate event.

D

Okay, all right, that's good! So then I think, uh looking at uh new relics, uh pageview timing, there probably is uh an overlap between uh what we have in timing and your pageview right. So that's, okay,.

A

So the the intent here- and I I don't know like if that's what we want to do here, but we actually have this event at new york as individual timing, like one timing, timing, piece of data, so it has a name and a value, and so it captures things like largest contextual paint or these paint events. It also captures window load and window unload and it captures even things like uh first interaction. It's like the time that the user interacted with the page or or when the page was hidden like we know.

A

If the user switched tabs, then we recapture it as a as a timing event.

D

So I I think I followed the everything except the first comment you made, which is one per uh you know. I don't think I'm following that. Can you explain that a little bit uh into individual event or something right so would you expect multiple pageview timings fired from a single page? Yes,.

A

D

Okay, uh uh what are the times when that will be fired, then, every time an interaction happens or for every object or something.

A

Yeah, so um so it's we um uh yeah it's. So it's it's for things that have you know, maybe maybe it needs to be cleaned up, but it's like we have um you know, so it does not capture like to just be clear. It does not capture navigation timing, navigation timing is its own thing.

D

A

But but things like largest contentful paint that you know like timing, information that come from the observer, the performance observer. They are not part of the navigation timing, those those are um those are represented as single events here and that's that's because you know they may or may not happen.

A

So, like you know you, basically, as you observe them you, you report them as individual events, and we also use this um for um for, like some significant or like some events that we um that we see as significant, which are like the first time the user interacted with the page, because we use that for the first input delay.

A

So, like the page view, timing, the we capture like the time that the user interacted with the page for the first time and as an additional attribute, we can capture the first input delay um and then we also have.

A

um We also have the page hide like the first time the the page was hidden and that's because we capture um you know cls is as an attribute um on that.

D

Okay, um I assume is you know, which is that a new category, you know uh event that we have to uh start capturing for from, or does it fit into any of these existing ones that we've already come up with, for example,.

A

I think it depends how you how you want to how you want to use the timing event, um so I think there's uh for timing. There is like I kind of envisioned having or I proposed, like you know like a few three weeks ago or a couple months, maybe a month ago, that having uh an event for navigation timing or navigation and having an event for resource timing,.

D

Oh, I see okay.

A

But then there are some additional timing: data that don't fit into those two so which.

D

Is what you know is that the category that you were talking about, yeah yeah.

B

Yeah they sound a bit like user actions, especially like the you know, the navigating away the visibility change and the uh the page unload, um uh and especially with that, like user interaction, because they get loaded up and it could be ages before they actually click on the page or move them out or something so.

A

Yeah, so I don't know I'd be curious to know how you capture web vitals right now, because that's one thing that we've kind of struggled with figuring out how to represent web vitals, um and that's that's how we that's what we landed on like basically um having them as like these these. uh These special timing, events that have additional attributes um that are specific to that web, vital.

D

Navigation timing and then um what was the other one that we uh resource.

A

Resource timing.

D

um I'm not really familiar with that. I don't think we capture any of these things in our events. Right now,.

B

Yeah, but by default we don't, but we have quite a few teams that do and they have their own um event.

D

Okay, yeah, I I I didn't make sense enough. There's a use case uh that we are already aware of and stuff. We can. You know, work on. You know defining that event. You know I'm definitely open to that.

D

So um I guess what I can do is. uh You know uh essentially create another tag here. Let's say we're going to try and define the that event. The web items timing event and we can give it a name and stuff, so we can continue discussing either here, offline or stuff. Let me um make one here just.

D

I'm just going to call it red white spoon will be here.

A

Okay, yeah yeah they're, like the weird thing about web, I listed like they're they're kind of each of the three bad wireless is different, uh like the the largest content for paint is just just the timing that that's easy, but then then, like the the cls is a score that keeps changing um throughout the throughout the life cycle of the page and and uh like first input. Delay is um is sort of is, is linked is closely linked with the time that the interact first interaction happened.

A

um So the timing is timing is the first interaction since the beginning of the page, but the first input delay is just like the duration that the page responded.

D

Okay, all right uh uh thanks I'll um I'll I'll I'll come back. uh That one will be continuing that. So uh you know continuing that uh list of questions um yeah for splunk.

D

Do you have a page view or an impression was the question I don't think I uh found it here.

B

I've used the document load span.

D

Oh yeah, so the link doesn't exist. I don't know what happened to it. Let me move it. uh So that's probably the scale if you want, I remember moving it. I might have killed it accidentally.

D

Yeah I'll try and get it back. Okay, that makes sense um yeah.

C

The google docs will have history, so you should be able to retrieve it so.

B

Yeah, I think remember when I was looking at you had a bunch of additional splunk attributes as well in there.

D

Are you uh you're not talking to me right now.

B

Not a teacher thing because I, which is why I added the the other two uh for two for two reasons. One is, uh I put them beside this funk one because spock was using them, but it also showed where they came from uh like one come from comes from the contrib um repo and the other one comes from the js repo.

D

I think I got it back here, I'll just throw it here.

B

And I thought there was someone else who was also using the documents managers.

D

B

um Sorry random thoughts, I I vaguely remember, I thought there was a another company that was using the document span. Loan.

B

Not seeing it now, though,.

D

Sorry I'll, reformat and stuff, so that makes sense so open telemetry um document load that dj that's yeah, okay, so I'll get these things in into that in a comparison, so that makes sense um url some of these things. uh These seem like um timing event. Also right. All of these things are timing, related stuff.

D

Okay, so it'll be split up into two separate events. The main ones are here: do you also have a concept of uh general by chance? I think you do right. Splunk, um I mean crowbar global event. um That's this thing! Okay, got it; okay! That makes sense so that clarifies uh you know. I think uh that'll populate um these.

C

D

Pretty pretty good, I.

C

D

And I'll you know post a message getting a discussion and stuff. So let me go over a few other things: app dynamics based event, uh that is, that the same concept uh centers the.

A

D

Is that the thing that exists everywhere, for you guys or the base page.

C

um No, but we do have a common set of attributes as well, but those I have included in this already.

D

Gotcha, so what is the base page then? The.

C

Base page is the the page.

C

It's it's the same thing. We call it base page to compare against uh the ajax index hr, but it's the page.

D

C

um We also capture iframes uh and to distinguish, I think, uh looks like.

D

That's the reason they have used.

C

D

I see so for a iframe, then you will not fire a base page either.

C

um We talked in some other context, essentially once we introduce like for the ajax calls, which is the page that is making that call. So once we have that attribute, then for an iframe. uh You know that attribute will represent the parent page, that included the base page and for the real you know the top most base page topmost page. You know that parent page will be empty, so so it will. It will still be a page view, except that there will be an additional attribute to capture. Who is the parent that invoked? You.

D

Okay, all right.

C

So yeah so in in short, uh even for iframes, it is the page view that will be involved that will be sent.

D

Okay, so that that looks like so that's the parent page, url uh concept here, right, okay, that makes sense.

D

Okay, I see all right. Do anybody else? Have a similar concept of capturing parent, page or filing.

A

D

Okay, all right, um so things like that, uh I think you know I I definitely you know, leave that to individual to see if these need to be promoted as top level fields in hotel. If not enough, there's two categories right: you can basically look at them and go. We can take an example say. Behavior is a thing that I can talk to all about. Behavior is the thing that we capture currently as part of page view.

D

Ideally, we would want to see that in hotel events, but if it doesn't make sense for everybody, then they draw on the you know extra attributes or properties that we can add to any event. um That's you know either that or if we strongly believe that.

C

What does it represent?.

D

uh Behavior is, um let me see uh the interaction behavior. You know, that's a that's a good question, one quick! Second uh nev. Do you remember we have a whole bunch of enumerations for behavior for page view.

B

uh I'm just gonna go and pull it down from the cdn and then I'll drop that link, so you can see them because there's loads of them thanks.

D

For page actions, interactions behavior make uh a lot of sense, but there's also page view related behavior that our customers capture. That's why it's there, um while we're looking at that, you know that's just an example that I took. If I, if we strongly believe that behavior must make as a top level field in order, then we have to make the case and get it up there. That's kind of what I'm saying establishing you know how we could you know normalize these things and create other events right, but if.

A

D

Like ah it's a it's, it's kind of unique to one company or something, then we can throw them as additional attributes that you know it may or may not be fully defined that people can capture.

D

I I hope that makes sense. So what what we would do is obvious things that have common across no questions asked we'll just throw them there anything else. Individual vendors should really. You know, take a closer look and say you know what uh let's discuss them to promote them as top level fields or by default. They go as you know, uh not going to be captured, and then you have an option of capturing those as custom attributes.

B

Yeah the custom attributes or a vendor specific um attribute name a bit like the the splunk ones, uh prefer splunk top well, it would be nice to not have the vendor name repeated multiple times. um I've dropped the link in there effectively it's the the javascript code, which um that effectively defines the enum. So you can see all the different ranges and what they are there's just loads of them.

B

Okay and they're, updated from time to time.

D

Yeah um yeah, so this is that's it's a huge list. Are you know um the the sdk does not really maintain this so to speak. There is a team that defines these things and maintaining things. uh I don't know if we want to go down this, but if people are interested in what this behavior is, you can look into it. um Sort of like a tendency of conversational thing.

B

Yeah most of these are used as part of the the page action or the the user action that we have.

D

Yeah yeah, that's great! um Okay! Switching back uh here um we talked about the base view and a base page global event. We talked about it, yeah splunk, post doc load. Can you explain a little bit about.

B

It it's basically same as resource fetch spends, but because the instrumentation doesn't include after base load happened, the when you have static resource on the page it the document load, instrumentation, doesn't.

D

B

Do the after added, so we added the simplification to also gather those in space and starch.

D

B

I I I think, it's probably equivalent to what uh martin was talking about with resource timings, so you want to capture the all the resources like the the css, the images and stuff like that, is that right.

A

D

D

We'll start load resource.

D

So, to be honest, I would merge them only into one instrumentation.

D

Okay um and nobody here from honeycomb, uh but does anybody know what a honeycomb just event is that uh the same concept of base event is that what um didn't look like it just says: target element, uh target element id and things. It seems like um page action or interaction to them.

A

So we can, uh I think purvi is from from honeycomb. I think I think she she attends the meeting tomorrow. We could ask okay.

D

All right, so we can ask them. Okay, those are the those are the questions I had um written down on things and um I'll continue to you know clean up the dock and move stuff around and try to populate this atmosphere as I can and reach out to folks on slack or these meetings to continue to pull out but like, like I mentioned, the intent is to get these columns filled up for hotel events at the end of it then we'll you know, then we can decide what the shape is going to be.

D

That's where we need to talk about whether we have tested attributes or you know. What mechanism are we going to use to track these events or send these events and stuff? I would like to separate these discussions. um I've been talking to nav about it. Also.

D

um My goal is to identify the fields now the data that we have to capture as part of these things, when we will capture them, what we call individual attributes and stuff- that's semantics, so to speak, and then we can, you know, as a group, we can figure out the transport mechanism or, what's the shape of the event going to be and stuff hope. It makes sense.

A

C

So at the bottom uh tabs, uh what did we decide on the resource? Do you want to have um one more entry for uh the resource timing.

D

Yeah that actually makes sense. Okay, so timing event. uh So we did talk about um sorry. um There's a navigation timing, resource timing and the violence so seems like we have.

C

So the navigation timing is for the page.

C

So so you can rename it to page timing.

D

Yeah, I agree any any objections on calling a page.

A

I mean I would I would vote for navigation timing because that's consistent with what they call it in in the api. But.

D

I see yeah yeah, I'm personally. Okay with that navigation timing makes sense, yeah.

B

Yeah and and as I said earlier, if it's a spa you'll have a page view without page timing, because there is no navigation event for the virtual pages, but navigation timing makes sense to me too.

C

Yeah, it's okay! What how about page navigation timing? Because we want to make sure it! It refers to that page view.

B

Yeah as long as we all understand what we're talking about, we could have the discussion about exactly what we call it later.

D

For now, uh and then you know if we have to reduce the size or something like that, we can work on, uh not you know producing it, but as a concept, page navigation timing, yeah, definitely makes sense.

C

And so you will have one more called resource timing. Yes,.

D

Are you something different because um apologies I'm still getting used to non-microsoft, uh you know excel, and you know the the sheets and stuff so.

C

Resource timing.

D

Let's just start with that, okay and then this webvite is timing. You know we can you know it's, we can name it whatever um a later thing that makes sense.

C

Okay, all right, um there is a page view itself. um We might um want to have two types of events. um I think one is to indicate the impression that the user visited right that want to fire right away when.

D

C

When the page, uh you know load at the first opportunity, but uh I also like the way uh the current document load. uh Instrumentation creates a span um because it it it shows a nice tree structure where there is this amount of time uh the page took to load and this much uh contributed to fetching the url. uh So the server took this much space and the remaining time it took to get the resources.

C

So we might want to have two messages, one as an event and one as a spam. The event you fire right away and the span you fire. You know a little later.

D

C

Because the difference is in the span, you could have a duration right. You would begin and end yes, whereas in the page view event that you fire right away, there is no duration. It it only captures that hey at this time you visited.

D

A

So you could in in theory you could still like assemble that that uh those at least for page page load, you could still assemble that from events from the navigation timing and the resource timing. Events, yes, but I but I agree that, like it's, it's convenient like if you're, if, like the if the end user wants to you know, have this visualized in like a uh tracing ui, then it's convenient to send it a spans.

B

Or to have a uh an overwriting spam like that, could be a separate instrumentation where you say on the sdk startup, it creates a span and then, when it's finished uh preparing so the point of I guess the first content will paint or maybe the user interaction. You then end the span, so you could use that and then the event to say page view could then have the uh could be linked to that span. Id.

D

A

D

Think that makes sense.

A

Yeah, I also want to also say one more thing is like so right now the current instrumentation it generates these spans after the fact they happen and it kind of links them kind of artificially, um whereas it would be useful. I think it whether it where I see the real uh like usefulness of this is, if you had, if you were capturing like the um fetch or xhr in real time as they're happening and like linkedin, then even the back end backhand spans.

A

So you could have like uh you could have like the the page load spam being the top level span with like, like xhr child child spans and then like the back end spans as child spends at that, um which is not happening right now.

B

um I think, but there is a definition where I think this part of the I might be wrong here. The document load event. It does go. Looking for the meta tag on the page called traceparent um to try and do that linking.

A

Yeah, but I mean like the xhr call, making a call to to the back end. No, oh.

B

Yeah and that's the case of you know when the instrumentation that instruments fetch in xhr, it needs to be a child span of whatever the current act of spain is which uh uh I still have issues with the way they get. The current active span from a global perspective.

A

Exactly so right now those do the child spend. The child spends of the document load come from the resource timing, yeah currently, which you can't you can't link them to back end.

B

So I think the document load today it like creates it populates. It then ends it straight away. That's part of the reason: isn't it.

A

C

Yeah yeah, I think the details we will need to work out. um I I think the um the the event that we fire right away. We still need to create.

C

We need to start a trace. uh You know for the span that we'll send later the.

A

C

For that span will come later, but we need to start the span so that we get the trace id that we include in the event that we fire, but at that time we should also take into consideration the the trace id that comes in the meta tag, jump from the page yeah.

B

um I I think, there's slightly two different concepts, but they are linked in terms of how we want to send them so yeah. I from my view, I think the the page span that you're talking about isn't actually an event.

B

It is just the the grouping mechanism, um but I agree we should define the the you know when do we create that span um and the sdk should should do that for us uh most likely at you know during initialization, because that's gonna be the first time we can do it, um but then we have to discuss.

B

When do we end it, um and that would be, which is why I'm saying it's an instrumentation um where it just goes and creates it and then how we inherit that down to the other ones without doing the global thing. That's still an open question. That's implementation.

D

So yeah it makes sense um the the span. You know the shape of the span event or the details that we want to catch it as part of that span. What would that um you know I'm still trying to separate? uh You know the the schema from the actual mechanism we'll be using. You know, you know: are we going to fight in a span or an event or whatever? It is right?

D

I'm trying to separate that um so, but I I agree, the timing of when you, you know fire these things and the linking and everything makes perfect sense. The tree structure is the thing that I'm going to be interested in is what is that span and what is the shape of that span um going to be? Or you know what are the details we want to capture as part of that?

D

Is that something that we have to define, as quote unquote an event, a schema here, or will it be one of these things uh that that's when I'm a little bit confused? Maybe I'm uh you know, I'm not making sense. uh Yeah.

B

I I think it's something different um where the events that we've got down the bottom of the page here, if you want to have those events included in the span, then there would be a span event, but the span itself.

B

uh I think we need to come up with a name, and then we need to come up with whatever attributes, which are probably duplicated. Like the page view, events and what's included in that span event are probably.

B

I think this is going to be significant crossover, um um especially for like when it comes to like the um well. We have your app dynamics of the url or page rehearsal. You'll probably want that in in that um yeah.

D

Okay, so um uh sorry, I think martin was going to say something I'll. Let him go.

A

I was gonna say like one thing that I um wanted to point out also is is like, should we have a should we have a section for uh attributes that we send in all of these events that are common and they're? You know so like the page url, for example,.

D

A

So I could shouldn't that be included with any of these events and spans that are generated.

D

Yeah, I agree, so I think we would get to it once we define these things. All the hotel events are defined. The things that we have to capture are defined. Then we put them side by side and basically go. This is common across everything. Let's just abstract them and put them, as you know, must exist in every event.

D

B

I I think we define the names first, because something like page title and url is very closely linked to the session id. So if we can get the ephemeral resources in there, then it would be really nice to put it up there and that way your common set just are there by default.

A

Yeah and like, if you then then like it wouldn't matter, if you well, then those those attributes would apply to to like any event and span that you send right. Yes,.

D

Yeah makes sense: okay,.

C

um um So for the tabs, I I would still suggest creating duplicating the page view and for now in in parenthesis you add a page view event and a page view span.

D

I see okay, oops.

C

So in parenthesis, next to it you say.

C

And then one more duplicate that and then call it a span.

C

uh The the actual names we could refine.

D

C

We'll be sending this page view twice once as event and once has span.

A

So do should we have also duplication for the resource and and fetch.

C

Or can you clarify that I didn't follow, why why do you need duplicates there so.

A

So um the current document load- uh you know you were saying like that. You want to have the relationship between the the xhr course and the document load span like the tree.

A

So currently, that's that comes from the resource timing.

A

Like the xhr calls and.

C

So the xhr call and the parents parent page span. They are not part of the same trace, though.

C

Okay, um it applies to resources as well.

A

They they are, they are if they, if the resource happens during document load, then it's taken from the resource timing.

C

Okay, but if a resource is loaded much later like in a you know, twitter feed where, as you scroll, the additional resources are loaded. Those cases the resource span, resource timing span will not be part of the page view trace. That's.

A

C

C

So maybe in that case, um should we completely take it out out of the document load trace and keep it separate, irrespective of when it is loaded.

D

uh Document load trace, I I don't think I'm following maybe enough.

C

Yeah, so so the example ram is that most websites, they load their resources like css, javascript images right when the page loads, but there are many other emerging. You know recent websites where, as you scroll down new content, is loaded.

D

C

Right so, in that case uh the page view event is not filed uh when you scroll, when you extend the page, but new resources are loaded, so the resource timing, event resource timing message. Let me use the term message, source timing message could be sent anytime.

C

You know, while that page is still in the browser tab, so in in one case, uh the resource timing could be part of a trace that the page view uh the page view span. You see the second page you spend part of one sum trace and the resource timing span could be part of the trace and, and also second case is given that it can happen much later.

C

We could still add it in the same trace. uh That's not an issue or it could be. You.

B

Know once you close the span, you can't add anything else to it. So once you end the span.

D

The scan is going to end uh when the document is done right. So the first you know first impression that you see um the page we span would have ended.

B

Yeah and nothing gets sent until you end it that.

C

Is not entirely uh uh if there is a there is a slight uh difference. uh Nav, I think uh a trace never ends a span ends right once a span ends we're not adding anything to the span.

D

C

Just because the span ended doesn't mean we can't add a sibling span in place. Yeah.

B

Yeah, which is what I see that so I I get a couple of comments first, uh so I think the page view spans you probably should either call it document, load span or page load span, because that's really what it is um for the fetch and the xhr, I think, by default. They will have spans anyway because they're asynchronous events, so it makes sense that they would start a spam and a spam.

B

The resource timing for those particular ones are not available at the point of the xhr on the fetch call finishing, you have to wait and poll to get the data um so keeping that as a separate event after the fact would be nice, although today we actually keep the the equivalent of our span, I don't send the event until I I've actually got the navigation uh timing for that event. uh For that request, or um I time out uh polling for it.

B

So I think it really is. The resource timing could be a separate event on its own and it may or may not be linked with the span of the fetch, xhr call um and having it defined. As a separate thing saying, this is the resource timing thing, for this thing um means that the url could be the css, the image as you're scrolling down the page. The effect it'll do a fetch or an xh out of to fetch the data and then upload the div. So I think they're all related.

C

I'm thinking um for for for easier representation should we capture this uh in the form of a diagram where, in a tree, diagram where we can relate, you know.

D

Yeah, I I I was- I was thinking in my mind about that. Also so, like a um a network, uh you know call stack or something that you would see right, yeah that.

B

D

uh I can I can try and put that some in our main dock and find that.

C

Now I think it will be good to uh um do. We have right access to this excel yeah, yeah, okay, okay, then I can.

D

C

Tab and uh I can create something.

D

Okay sounds good yeah go ahead and do that just just let me know when you know uh you know: significant changes are done or something so that otherwise, no um I'm not going crazy. But you know when.

A

Did I had this.

D

Again: okay, yeah cool.

A

I also want to like really quick uh uh point out that, like we did, we did a proof of concept like six months ago or so um with t2 and and uh martin worked on that too, um where we, um where we do actually have visualization that shows uh different spans and events in the same ui and they're based on it's based on the it includes the current instrumentation.

A

That's that's in the contrib, and it also has additional instrumentation for events that we thought we would be capturing um again. Let me send you that link to that, um but we can. We could also like if it would help we could modify that uh proof of concept to uh to help us.

A

Like kind of think of this.

D

B

Yeah that used to be listed in the top didn't that no work in progress, I'm not seeing it my browser prototype, that's the one.

A

Yeah I put a link in the in uh in the zoom zoom chat.

A

So so you can you can here, you can like invoke different things um and then like to to like trick to trigger instrumentation, to generate the data. So there's like maybe hard navigation, self navigation uh xhrs and then, if you click, oh that's, that's probably that's. If you go back.

A

And refresh and then open the there's, a button on the top right, open, viewer and then click on the timeline link.

A

And then this is like the visualization and the the things that are they're. Just dots, that's events and then things that um those are spans. And you can. You can zoom in on that. I think by holding shift and and.

D

um Maybe um maybe you can share and show us. I think that yeah yeah so interesting so am I.

A

B

D

Thing, uh martin.

D

And you would expect it to expand uh these things or something yeah.

A

Yeah yeah, if you, if you, if you hold shift, then like you scroll with your wheel, then it should like expand. Okay, yeah.

D

A

And so the ones that have they're filled, those are spans that they actually come from the existing instrumentations and then the ones with dots.

D

A

Those are proof of concept events and this this was like six months ago. So, like our you know, we have, it probably would change to some extent but um yeah.

D

Yeah this makes sense. This is kind of what uh you wanted. One of the right uh centers.

C

Yeah, um how did you open this sorry, I? I was also trying this out um yeah.

D

C

This window, martin.

A

There's there's a there's a button on the top right corner in that uh first window, open viewer, yeah.

C

Yeah I, when I click on it, I get a 404.

A

It's probably you know, so you have there's a bug in there. Probably so you have to have. You have to make sure that uh the url is what uh ram is showing right now.

C

D

And then um click open viewer. Okay. I got it now on the timeline.

C

C

But here uh this is a session right in the session. What all events have happened, what out spans and events ever.

C

Right right, yeah, it's uh something uh similar, uh but but I think uh um yeah, I think it's it's the same thing then yeah.

B

So these are all effectively just spam events right looking at them.

A

No they're actual log events, they're they're logs.

A

D

A

So we can, if it's helpful, we can, we can expand on this. We can modify it to so. It sort of aligns with what we're working on.

C

I I uh I was thinking more of a static diagram, just to illustrate that which type of message know relates to how they it's a relationship diagram. Basically, you know how uh one message relates to the another.

D

C

In some cases, we're talking about hey this being part of the same trace as another. In another case you know they could be unrelated. So that's the part. I I wanted to capture okay.

D

A

I think it sounds like we should probably have more more discussion around the resource, research events and spans.

D

Yeah definitely, um but I think you know um what wanted to get is all on the same page about you know what we're attempting to do and stuff in terms of things you know I'll continue to. You know, try and separate the action implementations from the schema itself.

D

Sometimes we we talked about them and you know um so try to intended to separate that, but um yeah, I think we're not uh up to a good start. um You know, hopefully we'll have these events defined soon now the schema is.

A

Defined soon so.

D

All the terms are overused events mean different things: yeah schema defined, yeah cool thanks, guys.

A

Yeah thanks for driving this ryan. Well, no.

D

Problem yeah, okay,.

C

D

C

D

C

D

B