Open Telemetry Uncategorized, 4 Oct 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2022-10-04 meeting

Description

Instrumentation: Messaging

A

A

Welcome back hey good to be back.

B

Have things been these past three months.

A

uh I think there is some progress on the lock side, the logs API was merged. Oh good, the events API.

B

Great, that's awesome. I'm, really glad to hear that yeah.

C

Complete with defining attributions attributes, not nested attributes.

A

There are some things that are dragging on.

B

Yeah yeah no no shot no shock there. Yeah and.

A

On the on on your topic of ephemeral, attributes, I did see some new discussion uh between tigran and Josh, because I think for metrics.

A

There is a requirement coming up where not all resource attributes contribute to the identity of the metric. Yes, and therefore they are thinking of having another set of attributes on the resource. Right, we have an opportunity to.

A

um You know uh called some of those attributes. You know as as something that would change uh call it ephemeral or you know, modifying or varying attributes, but I think that would be useful to us. Yeah.

B

Now, identity attributes is just like the other side of the stick, as it were right, like people don't want resource attributes to change because they use some of them as identity, attributes yeah. But it's like if you, if you can Define either group, then then you're, okay, it's just when you don't have a definition of an identity.

B

Then it becomes scary to to have some of them be changeable. So that's great yeah.

A

uh Never do you know if Ram is Johnny.

C

uh He did plan to when I was talking to him yesterday, he's currently showing us offline on teams. Okay, um sorry something may have come up.

A

We can do it ourselves too, or we can get started.

C

Yeah because in our daily stand-up yesterday he said he was going to work on it. So.

B

C

Actually, the the spreadsheet might be good for Ted as well, because it highlights the potential ephemeral resources. So we came across a session, ID user, ID, authenticated user, ID I. Think that's something else. Wasn't there.

A

So Ted for context um what um um we have been doing and and Ram from Microsoft you know, has been driving this. uh What we did is we are trying to uh Define some kind of a schema for all the uh browser side, uh spans and events, um and we collected uh for from each vendor. We collected the you know existing uh definition of their events, um so, for example, for new delicate page view would look like this.

A

You know these are the fields similarly for each uh vendor, and then we put them in an Excel sheet and we we yeah the First Column is, is the common set of the common attributes that are applicable to all the vendors? That's great! That's awesome, yeah!

A

So now that Ram you're here uh will let you drive. I had one uh items uh on the Ajax event from last week.

D

Nice to see you welcome back, hey.

B

It's safety man, yeah.

D

um Cool uh thanks for the intro and apologies for joining late, um I was going to say you know this week is probably a little bit uh tough to spend time on this. um We're closing on. You know, planning for the next semester and stuff um some presentations to uh um leadership and things so anyway, I'll try to spend as much time as I can um yeah Santos.

D

Do you want to start with uh your agenda item and then they'll probably give us all time to you know, get back into into group and start going through yeah yeah.

A

I think for the uh you know, this is the tab for.

A

uh The xhr and fetch correct, um so we want to capture uh the page URL of the page uh from where the calls were made right and that I'm thinking will add a new attribute uh it. It doesn't exist today, uh so I'll add at the bottom yeah.

C

I'm pretty sure we've got it somewhere, yeah.

D

I'll, stick it at the same time.

A

Is it this app Dot Page or your page URL um for for an Eco, no.

C

I think it's probably one of these uh tags. So let me look at the code.

C

um So we have, these tags are like AI, Dot and then, like operational ID is like where we have the session ID and stuff like that. Yeah.

D

The operation name, if I remember correctly name, is the is the target uh right that we're calling.

C

Yeah um I thought that was the URL at the top. Okay.

B

C

Let me go find the code rummage through it.

D

While Liv's looking uh perfect, can you confirm uh for honeycomb um the thing that uh app.page URL that's what this is pointing to? Is that uh used for that purpose? For you guys.

B

um Yes, okay, the.

D

B

D

What we need to see is for AWS are: um do you currently capture that uh centers.

D

Doesn't look like.

A

It yeah, but I think we do um maybe I missed it. I, okay, yeah, okay, I think it's called I think it was their.

A

D

I think that's okay, but it makes sense. uh It's a useful thing right.

A

This I'm thinking.

A

And um the other part uh here is that this attribute um does not change. um While you are on the page and when you and and therefore it could be an attribute on the resource, and when you navigate to a new page, your SDK is going to reinitialize.

D

It's it's! Yes, if you assume it's a regular app, if it's a spa, then the page URL could potentially change.

E

Which is to say, the history API to solve changing the URL.

A

So first single page applications.

A

So so we we want to allow it allow this attribute to change.

A

Okay, in that case, uh it will go into the.

D

Yeah I see you fill it out. uh This is cool thanks, I think you did this right, satosh the two tables that, after we talked about last minute. Thank you.

A

I'll I'll extend this more later. Yeah, okay and um I had a question on this transfer side transfer size that we added last week.

D

um Did we did we add it last week it seems like it's somebody uh I, don't remember talking about it. I.

A

Said: okay, okay, I saw this here and uh I was trying to understand what it is and I noticed that it is. It is currently um by the performance, timing apis, um so it doesn't have to be a high level attribute. It could be one of the timing entries.

C

uh Yeah I think all the ones here are timing entries and um one of the comments I had there for the from my I think from last week is all the green bits. The performance timing defines effectively the start and the end uh our Ajax Perth stuff is effectively a collapsing of that. So it is uh the duration of each component. So my question here is: do we want to have and transmit the start in the end, or do we want to do something similar from an Hotel perspective?

C

So I think one of those links is to the appetite's code that has a nice, the last yard graph of performance timing mappings. It's probably the second one.

C

Yeah there we go so these are effectively all the performance timing from the API. um So the start time redirect start redirecting Etc or all the performance timings. And then the bits at the bottom, from like 81 to 84, is now versions of those.

C

So rather than sending like all 12-ish of those, it sends a smaller set.

E

uh It doesn't I think this chart is missing the gap between connected and then to get start, which is basically the queuing time.

C

uh Well, this is specifically for Ajax requests uh not for the pace mode, but.

E

You can also have queuing great text requests too.

C

E

C

Code's really old, so yeah when this was created.

E

So I'm saying this is a case for keeping it as that. Browser then gives the.

C

Data yeah so when as new one gets get out of, they just get automatically included, uh which means we're saying the back end has to. If I could do these calculations.

D

um One uh quick thought: if we did that, um would that uh with that expander would be amend the schema every time, uh the performance timing, um spec. uh You know for the browser's change.

C

um The way where we did when Martin uh did the performance timing. Originally uh we proposed using wording of uh it would follow the w3c standard. So when, when they updated it, we effectively automatically got amended, got.

D

It, uh the okay, I I, think that makes sense, so basically that uh the we would start collecting new metrics. If you will from the performance timing, object as performance timing object as it becomes available. Yeah.

C

Because they're, not always there so you've got all browsers support all of them and like the domain lookup, it's only is it only present if okay, there was a DNS entry that that performed as part of that request. Okay,.

D

So because we we're going to offload this, this processing of switching it back together or whatever it is aggregation, you know whatever you can think of right to the back end. Do we need to specify some sort of a version or something like that along with that, or do we just say, backend will say if that property is available, then they use it. If not, they know that this is from an older browser. Something like that.

C

The um no it's probably more a case of we'll say this: this is the defined Set. uh You may get extra.

B

C

um And the back end can determine whether they just drop it or not, but it will be sent as part of the um the protocol.

D

Okay sounds good: okay, okay,.

C

D

So then, the green stuff is the defined set that we're going with correct. Oh probably, not sorry, I think I got confused, so the the green set is where we need to define the the initial set of defined things from.

C

Yeah, the first lick I think, is the link to Martin's doc from memory. That's what I put in this, uh which already defines a set of performance timings, that of which these resource timings are a set of those.

D

Got it so we need to someone needs to take that and plop it here. Basically, yeah.

C

Okay yeah, so these are the.

A

So um I'm I'm thinking um the fetch object and the fetch timing object uh can be two independent and related objects, while defining so that yeah yeah so I either I can open a new tab or in the same tab in the bottom. I can have a separation and then Define the timing.

E

I, wouldn't timing, questions just be resource timing, tab.

C

A special talker, so we'd have to have a mechanism for linking the fetch resource timing, with the fetch call itself.

A

So the timing entries are, they are they're same for resources and uh patch and Ajax.

E

But fetch has a resource so where you get those data.

A

Okay, then, then, the all the timing info we can capture in the resource timing, then.

C

uh Yeah and then we have a a type field, that sort of said whether it was the fetch timing, xhr type or whatever, whatever we're going to call it. So the back end knows that it's related to that, because the the fetch timings aren't available immediately after the fetch call. So if we could send the log entry the log event immediately after the pitch call that works and because we have the time stamp being the time, the event was recorded.

C

um That works really nicely because it'll be at the time that the electrical came back.

A

So so you're saying we can have something like browser.com.

A

And this can be fetch.

B

Or which are resource, something like this.

C

um This is probably defined in Martin's dock yeah, because it's yeah, it's pitch, xhr, I, think resource or the page.

C

um It's because the yeah, the browsers, have the subset. So it was like there's a whole bunch of stuff that extends resource.

A

Okay and then the individual entries you know, go here: okay,.

B

C

Which is really just captured from the w3c specs for them all.

A

So these columns in this table- those are the types the types I see: okay and then we'll Define, all the different entries but which entry is applicable to which type you know that would be a different mapping. So then, do we rather Define them separately.

D

Can uh what what do you mean sorry, Santosh.

B

D

When you say that uh Define them separately, yeah.

A

So that uh we are precise, that for navigation, you know these are the entries that one should expect.

B

Oh I see: let's.

A

Say if there is no start time for navigation.

D

Okay, you're thinking about how to capture this in the Excel for the even scheme as probably separate tabs for each one uh would be helpful. Is everybody thinking.

A

A

I I was separately. I was reading um um a paper from Facebook about a schema first uh Telemetry, um where uh I think Facebook is going with the approach of predefining, all the schemas for all the telemetry and then using them right from the source. So when, when you're generating the Telemetry from the agents, you actually use those definitions to to create the Telemetry objects.

A

um Sir, so you know that all the data is going to be confirmed.

A

um So if we were to suggest in in the interest of uh aligning with that, it'll be good to be very specific. That, okay, which type of event, uh has exactly what uh attributes What fields.

D

Okay, I, don't disagree, I think you know largely that's what we're trying to do. I think you know, at least from you know, a Microsoft perspective. We've been in scheme of us for a very long time. Yeah. um You know Define schemas for everything and then custom events are always free for all type stuff. Even that we try to Define schemas.

D

You know um guide people to Define schemas before they produce Telemetry, because if we don't do that, the back end you know becomes, uh you know, lost essentially right now they just don't know what crap people are dumping into the back end and things um anyway, um I I agree. It makes sense.

B

D

I think the challenge is: how do we get this this table uh into our working dog? So we know that the navigation uh uh timing, I'll keep.

A

This table, as is yeah I'll copy, this table and uh I, think I'll. Remove these vendor names and I'll keep that table here.

D

Okay, you and I can probably work offline to see how we can get this into the Excel. Also yeah I, think that makes sense. Yeah thanks foreign.

A

Sorry one last thing on the Ajax: is um this uh row 17 and 19 I think they don't exist if it was added by mistake, I think we can remove it, but just wanted to bring it up. Whoever had.

D

It almost 79, we did talk about them so because description is filled in this is something we did talk about as a group and we added them.

A

Yeah there's only key and then the value goes in the attributes value. So there is no attribute name with uh with with this. In this format,.

D

So attribute doesn't exist separately. The header value will be the value of the corresponding attribute of the header represented by HTTP.

A

Yeah so let's say content length uh you want to capture content length header, then it would be http.request.header.content length equal to in order that number.

D

A

So you uh we can, we can just delete this I.

D

See yeah, it makes sense yeah.

A

But coming to um this cell Microsoft has request headers.header value. Yes,.

C

Because we have it as a um the nested object, so.

D

I think that's fine! You can ignore that I see it I see just just uh you know empty out the you know, 17A 17b.

A

B

B

Okay, that's a.

A

Good point all right: um that's all yeah.

D

Okay, um just as the normal exercise right do, we want to quickly go through uh if we missed anything from vendors, at least the people that are present here. We can do a um in a filter on that vendor, see if anything, pops up and we can discuss about them.

D

You know um just go to I mean Cisco. Do a filter. There remove links, foreign.

D

B

D

Let's do that for honeycomb.

A

There are a bunch of things here.

B

Yeah I think I, don't think, there's like necessarily apply to to I guess to like fetch timing, specific stuff, I. Think that's something! That's more like metadata that we pack on to most of our our spans and yeah.

A

So these um all go into resource, but HTTP route- I, yes, I'm wondering how is it different from URL.

B

um I think it's just the actually I think that should probably be path.

B

um I, think we call it route at the moment, but it should probably be called like just the path.

B

It doesn't necessarily need to be separate as long as the like we get the full URL and but I think it would be. It would be interesting to think about having like the URL, potentially split up into different um important pieces like being able to just filter on a path. How do people feel about that.

E

This can be hit on post-processing server side, because that does mean duplicating duplicating bytes, for instance,.

A

Url has like a number of components, right yeah. So if we miss anything, it's it's going to be confusing.

B

Yeah, that sounds good, so as long as it's captured.

C

B

The URL I think that's good. Okay, yeah.

C

The cardinality of the full um URL is always going to be painful too, uh although I had like lots of internal teams complaining and wanting to strip off the query strings because of eudb requirements.

C

um So you know pii concerns uh we're going to have to have to Define it. As you know, depending on the situation, the instrumentation may only capture the uh the domain.

D

So that actually brings up another point. You know for page view for the URL we we set it to be sanitized and things do we want to put that description here too? um It should be sanitized right.

D

Ajax, most likely will contain a lot of piis. um You know the parameters are passed in as opposed to page the opportunities of leaking. A lot of data is high in terms of Ajax.

D

uh Put it in the description, yeah just copy the description from uh I can do it on that page yeah uh and then um just real quick. uh Just you know, uh tool wise right now. If you filter honeycomb again the fields that we have already gone through and what previous basically said, I don't want them in the main schema change them to um idolics just grab everything from 64 through 71.

D

got to like uh not not the first one, 663 alone yeah, just control, I cool. We can move on to Microsoft.

A

The success and type.

C

Yeah, so success is really just a Boolean if it's a 200 week that and type because we reuse uh this for Ajax or sorry for xhr or fetch, um but just to find that title.

D

We probably need that right um because we said: let's see, do we need that or yes.

E

It's in italic, so we already talked about it last time.

D

We didn't talk about it. Okay, the today we had the discussion about resource timing, stuff, so I'm thinking, maybe I'm confusing my head on how that's came out. Look it up.

D

um Where will we capture the type uh the we're going to have the resource timing thing right.

A

um So scope name is what we are going to use to determine. The type right okay is: is that a common agreement.

D

Scope name from sorry and spacing element.

A

A

D

From the span I see.

A

Yeah so let's say so, there is a scope.

C

Well, that defines the um the specific instrumentation, though not necessarily the type of request that was done. Yeah.

A

So we add something explicit.

C

I I, don't know how the type is used today, so I I, don't know if it's really required.

C

No, not because they're effectively an application is either going to use exit Tower it's going to use patch. So, apart from knowing that this was a remote dependency call I, don't think they need to understand to know that what what the API was used to send that Ajax request.

E

Even if there was a difference, I don't think it actually matters that much really yeah.

C

Exactly, which is why I I I I, don't think we did it. Okay,.

A

um Actually, um at Cisco 2, we just call it an Ajax event. Today we don't differentiate, but uh given that what we are defining here is is going to be for everyone yeah for everyone, I'm sure you know somebody is going to bring it up in future, so we might as well add it.

A

You you need it even for the page view span right. How do you know like what, like you in the back end, how how do people identify uh the spans? What identifies a span to be um a certain type?

A

Is it a DB call, or is it an HTTP call I think they look at presence of a certain attribute, um so here what would that attribute be, and if that's confusing, we might as well add a an attribute.

C

Yeah, like we're going to have the event names, the event name is going to be whatever the event name is that we come up with it for this.

C

But do we really need.

C

I know we could have it as an open question, because, whatever it is it it's probably going to be optional, because I I can't think of any concrete example of why you would want to know.

A

How would you differentiate between a page view span and a jack span.

C

Well, in terms of page view, event or an Ajax event, they're going to be different event, names.

A

So page view we said we are going to send both as a span and as an event, even if I right away as a span once the timing information is available.

D

Would be not be able to in the the page view and the page view the page view event at the page view span are going to be connected right with by the ID. So isn't that how you would get to all the spans that relate to the page view that particular page view or page views.

A

So you would use that information to no an event would have a reference to the span. A span would not have a reference to the event.

B

A

E

But it would kind of a reverse reference, because you can look up the event by the span. Id then.

A

Yeah, it's hard, though,.

E

But uh for spends there already is a span name called document resource. It could be exactly the same, but uh but the document the load spend name. It could be the same with a span name for it. So.

A

We can standardize on the fan name. Then yeah.

E

And then for all of data because speed resources or fetch theoreticals It's Kind, it's kind of standard for recording it. A message.

D

Okay, okay, a quick question: I thought: I thought the span name was um the actual um um I'm just trying to look for an example. Right so and you know forget, uh forget clients and everything your span name would be for any operation. You start uh you, you give it a Spanish name right, but what we want is a higher level concept which simply says this is a fetch. That's it. So if you fired it or I fired it, it'll be fetched, it has to be fetched, it cannot be.

D

uh You know, shopping cart called one, and then you know or something like that, uh vendor specific. So it seems like we're using. We might be using span name for the incorrect purpose here and we used it yeah. So.

E

I just uh linked the spec for that that matches the HTTP client span names and there it's critically. It says that all all everything that does a remote called over HTTP name TCP method, unless you can get the less Cardinal name for it.

A

Yeah I think that is what it is today. uh The name is HTTP get yeah.

B

C

So yeah so spend name is not no.

A

No, no, but but here the guidance to name it, this way is only to avoid a high cardinality. If you end up using uh the operation name, let's say the shopping. Cart uh uh is the operation. uh If you let's say you know, we use some part of the path uh uh in in as the name. Then it becomes high cardinality. But if you use a fixed name, then then I think we will still be compliant with this pack. It will be, in fact, a fixed cardinality.

E

A

Yeah, so so it is okay to use uh span name as a fixed uh like enum only these uh values it can take the page view. Xhr fetch the sun Beacon things like that.

D

No, since we're defining fetch timing, I think we should just specify the types that we could have right: it's basically it the types and what in stock right there.

E

Very for fetch, you do have a conflict. That span name is meant to be for recipe calls and at which is a HP call, so it would spend name should be HP get or post or whatever yeah.

C

Because the the actual part, the effect of the xhr, could be a it or a post.

A

But that will be part of the method right, the name you it could be anything and they just as a guideline. They, the spec, gave a suggestion. We don't have to follow it.

A

I I didn't uh maybe understand what you mean. I think.

E

I think in this case we need we might need to either modify the current spec or ask for clarification if it's meant for, like all HTTP calls or a safety for when it's not, and you don't have any better.

A

D

A

Get is too generic right and get account. Slash. 42 is too specific, so the the first row conflicts with with this recommendation.

E

But it does also have ACP in front of it, so it.

A

Says it's too specific I mean it's too uh generic generally.

E

A

E

Yeah, but it's also meant to see that if you don't have anything better than that, since, if you're interested all of the HP or someone will generate it's going to be all sorts of garbage that you just can't normally it's just like it is in this example, because you don't have clear routing information around to your client side.

C

And potential PIR yeah.

E

B

E

Still I'm doubting that the IDS are bi but yeah, there's no clear guy doing it anyway. It.

C

Depends on the application. Some applications do use the URL specifically so.

A

But if you use the type um as the span name, what is the concern.

E

If the fact that this fetch is a HTTP call, conflicts with the ACP span, spec.

D

Okay, yeah: that's.

B

D

So it the the fetch timing. If there happens to be HTTP call um the Ajax, then we would be using an HTTP span to capture this whole thing right because we agreed we will capture all of this data using spans. That's a mechanism which we said we will avoid talking about it, but we have to uh you know a little bit I think if that, if this is sent out as an HTTP span, then we'll be you know they actually, the span dot name should be in compliance with the HTTP span.

D

um Spec! That's! Let's do this point yep.

A

Which, which it is, um is what I'm trying to say is the spec um gave several possible examples. I think this suggestion is only to avoid uh yeah and in our case, by calling the span name as fetch. It is, in fact, a constant cardinality.

C

um Are we talking about circles because realistically we're saying we're going to have an event which is the fetch details and potentially an event which is the timing, details.

B

C

Therefore, like for fetch xhr, it makes like 100 percent uh sense to have a span so effectively before we even start. This we'd actually have a span that gets started as part of the fetch event. It would be linked with that span ID and then, as part of the fetch resource timing. It should also be linked with that span ID and then, after we're done, it would then close the span. So therefore, we could just use the span ID to link it together.

C

And the span will have its own name, which would be you know, I.

A

Think it could, uh um we could simply have something totally different too.

C

Yeah yeah, we could.

D

uh Would you call it ajax type or maybe.

C

A request type if we can yeah and potentially we could reuse that else right, so we could say: request.

A

If we were to use an existing namespace, would this work I think focus on the non-browser world will have concerns.

C

um Client types are very bad, I think is it that could infer a resource that says what type of crime you are: um okay, but yeah yeah having some other things yeah um and we can okay I'm sure, but once we put this out as the main spec we're going to get like lots of feedback on things, yeah.

B

A

Okay, so we'll move on.

D

uh Just make type a normal one, just just for it's.

B

D

Still have the filter version seems to be there properties. You can ignore that uh just custom properties talked about that Ajax. All of this stuff is talked about.

C

um Yeah, so this is potato all over the AI dot, pegs yeah. These are all the I think the potential for femoral resources, uh except for maybe the SDK version. That's not going to change nice, um but yeah. Most of those like the session ID, the user ID. The operation ID, would potentially be change. Changeable objects in a spa. What.

A

Does operation ID represent.

C

um It's sort of like the session ID.

B

D

That uh is that true name, you know I thought the operation in operation ID, where the the targets that are calling right. You had an ID for that and uh the name of the operation that you're calling the remote thing right.

C

um I I think we actually use operationid as this ban ID.

D

Yeah we do, we do so I. Think oh yeah, operation idea is a span ID. So that's the you know uh generated good of some sorts. The operation name is the remote thing. For you know, if you're calling my backend.com, um you know fetch shopping, cart or something like that. The fetch shopping cart ends up being the operation name.

D

That was my understanding. You, you probably know that uh yeah I.

C

Look at this occasionally I, don't look at it all the time um but yeah the session. Id I think when I did all the new distributed pricing refactoring. That was where that one landed um and it's different between the two code bases. So it doesn't help.

C

Let me go find operation, I. Think that's interesting! For the page, one uh page and email I found we have an internal variable, that's never populated, so I think we use the operation ID to link them on the on the back end.

D

The HTTP route is similar to that. um You know this. This thing that we're talking about um operation name. You know it's basically a piece of the URL in a component of the URL, that's operation, ID, if I remember correctly, I mean operation name, not operation, ID, qualities,.

C

Why is it under pressure? You can never find the piece of code you're looking for.

A

C

The effect always AI dot or actually sub-objects of tax. That's why tags is in um metallics.

A

D

No, we don't need tags yeah, we already talked about it. It's um at least we can ignore. Okay,.

A

Okay, I think that's it then yeah.

D

Yeah, it's its components: hostname Port are components of that. Url I, don't think we you know we would decided across the board we're not going to do that right.

A

Foreign body size, I thought we saw.

C

Yeah, that's part of the timings so.

A

Request body, is it or is it response.

C

I think it's response, size and transfer size of the true.

B

Request body size, I.

D

Just made the hostname import uh I'd like to just letting you guys know.

A

A request body size, I I, don't know: do we is it possible to capture that.

C

That's right in the wrong code or.

A

I can add a placeholder to check with Martin on his back. There.

D

Is a there is a size field uh in Martin's uh um a table.

D

Navigation resource it only applies to largest contentful paint and other things yeah, that's the only thing it applies to.

E

Every sorry they are an attribute for this in not respect too.

C

Okay, so finally found it once I started. Looking the right code base operation name exactly is the document location, sanitized.

D

uh Can you say that again document uh what.

C

Document.Location, so the URL that send it hi.

D

Yes, okay, a standardized! So that's that's a duplicate of uh what we already have right. So we can ignore that okay,.

A

What what duplicate of what you are.

D

Yeah, okay, I'll I'll, mark it there you can uh we're gonna continue with the current um request. Size thing.

A

Yeah for New Relic.

A

I think if we are both working on this.

D

Oh I see so: okay, somewhere I should say it's just for me. I think apologies yeah go for it all right! Yeah.

A

I think there are two remaining um time to load event, start time to load event and I I, don't know what they are.

B

C

That would almost be like the spine duration, but it's not nine times.

D

Time to time, the last last, oh.

A

So this is for the Callback when.

A

C

My last callback, so if I've got listeners or something that would make sense.

D

um Martin said we can, uh we could ping him on uh slack and he'll respond. uh It seems like in a vegan Channel. Let's, let's do that. Yeah.

B

So I think okay sounds good. We'll come back to that.

B

D

So so far we have um oh I, think Splunk. This hotel should all be captured, I think already.

D

D

A

Yeah I think we'll be good yeah. Okay,.

D

Cool just to you know we get. We got another seven minutes, I think just to quickly recap. We got the page view largely done. uh You know, you've transferred that uh over to the main doc on what the event needs to be and stuff right um same with fetch I believe so. In parallel, we agreed um people could start looking into how we you know we're going to implement this correct.

D

Yeah, as we continue with the other events- and you know, continue to Define them and things.

A

Yeah I think the uh like, for example, for page view, um we said we're going to have one event and one span yep.

A

uh So these two do they represent these and and then uh the navigation timing would be the corresponding timing object attached to the span, so the page view would be the one that gets sent out immediately when the page is loaded or when the page is visited at the first opportunity and then the page load span is when the loading completes through whatever definition um you construct the span um and then attach a timing event.

D

So um this is where we do. We, we Define the page navigation timing. We wrote down navigation timing, resource timing, user action Ajax way back when we just started right. So, let's revisit that, you know help understands the browser, slash page navigation timing. Is that a separate artifact, if you will uh or is it going to be part of one of these I, believe it's part of.

A

I think that could collapse I think uh like for each of these. We said we're going to have.

A

B

A

Have a common event where the type determines, what type it is and then yes um for each type, the the the attributes you know differ, so we could potentially collapse uh these two into one yeah.

C

And probably just a revisit your definition in terms of ordering I would think we would actually create the page span. So start start the span.

A

C

Could then fire the page view event? So it's linked to that span correct if I were any other events that we want to have, which would be the the navigation and all the resourcing and then once we're done. We would then end the paper, the page span and that um I think that would probably be the ordering of things so that, from a waterfall perspective, you'd have the page load span um and then head of all the events that would be directly associated with that, because that was the span that the events were linked to.

A

um One clarification, though, so the um then the timing, events uh in in your case, do you want them to be sent out as independent events or attached to the span? No.

C

It's still uh still say that as log events not not span events, although we could what yeah, assuming we have a mapping from the log event to a span event, because we have this nested attribute issue uh or if we already Define them as not having their attributes. So if we could represent the the event in the same way for both then I guess we can have that discussion, but um I'm thinking, separate event.

A

So um there are um some of us who want it attached to the spam piano all right. So if we were to satisfy both, um we can either.

B

A

um Now I'm I'm asking uh whether we should uh initiate uh there was one discussion in the past about uh introducing a configuration on the trace API. So you would still create a span events, but there is a configuration on the trace API, which would uh actually create those events as log events. Yeah, so should do you want that discussion to be opened up, um because that could uh take its own time. Yeah.

C

I I think, let's continue defining the events, yes, okay um and then once we've defined the events then I think we need to within ourselves Define how or if the events could be met as span events um and if they can't, but it would be nice to how do we represent them. Sorry, I I think, let's just Define the set of fields, first yeah, which I think we're really close. Having.

B

C

um But yeah, my comment was really conceptually in terms of The Ordering of things: um how how I think that should go in terms of how they get transmitted? That's still, the opening question got it so.

D

The the auditing of events I think that's what uh Santosh did in the main tab, I believe or made a timeline that things yeah.

A

Yeah I think that's not perfect, though. Okay.

D

So we should probably go look at that or maybe we should just write it down or something like that. We need to capture it. um That's a it's. Probably an uh an action item. Somebody's got to take and start writing it down on uh the waterfall like stuff right. What starts first? How do we attach the things at least a high level um and then uh real quick? If you want to go back to that main dock, uh sometimes where uh we're going to use that for implementation, uh the other one?

D

Yes, uh my I propose: we remove the page navigation timing from here. Okay, that's just confusing: let's remove it and there's one resource timing: let's uh you know, take Martin's table and you know try and ultimately translate that into the browser resource timing thing.

D

Does that make sense, yeah.

B

Yeah, okay, um so this.

A

B

D

Yeah, that is just confusing.

A

Okay- and this will.

D

This, this is what we need to somehow figure out how to get uh Martin's thing into a spreadsheet and here or directly here, something like that.

D

So uh we need, we need to work on that and then, as a group discussion, we probably need to start looking at other events. uh You know user action or interaction seems like so.

A

Will you be joining tomorrow around.

D

uh If you give me one second, I can I confirm uh just real, quick and I'll put it in there. If I can make.

A

It up yeah, no I, think before user action, I'm wondering uh are both user action and error. Error also is something that I, yes, yeah.

B

Okay, so then, then,.

D

I can probably start looking at error. That's one of the you know very fundamental things right. We.

B

Can do that and then go to user action.

D

Yeah I agree: we can start with error next: okay, cool.

B

A

D

Sounds good thanks. Everybody hopefully see you all tomorrow. All.

C

Right cheers. Thank you.