Open Telemetry Uncategorized, 6 Sep 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2022-09-06 meeting

Description

Instrumentation: Messaging

A

A

A

Morning morning,.

B

I just saw your comments um on the browser URL as a scope attribute.

B

Thanks for your comments, um I'm here to fully understand uh they're just there, but I think uh you know in the absence of the resource level, uh attributes that can change um so looks like you know you, you don't have any um objections to keeping them at the instrumentation scope level.

C

I mean as long as nobody else is going to be like No. This must be mutable. I can't guilty things. Yeah.

B

Yeah I'll I'll find out uh about that. um There.

C

Is also I think last hour there was something about attributes in the hotels, specific, okay.

B

About the scope, attributes.

A

B

Yeah I think uh yeah. There is a another discussion going on whether the scope attributes should be part of the identity for the uh Tracer yeah.

C

A

B

Good morning, nav.

A

B

Oh, we have Ram today.

B

Good morning, Ram uh I think uh we can talk about uh the events standardizing um topic since uh you are on the call today.

D

Sure um yeah Texas yeah.

B

So far, I think we do have a few folks uh um across several companies. You know add information in that document. You uh created sure.

D

um I, you know just a quick good thing: uh I was out on vacation uh in a couple of weeks and um just got back today. um That's the reason why it's not been any act in no activity at all there uh planning on uh looking at that. um So don't have a lot of updates from that point, but um and I apologies for just kind of going dark.

B

Uh-Huh no worries.

B

Yeah uh but um I think um like how do we, you know, proceed on that.

D

Yeah, so what I realized is there's also an Excel spreadsheet that um um a few of you worked on um early uh this year or something I think I.

B

Think that information is copied over here- uh okay, perfect, at least from um our side and Martin's side. uh We have copied okay, so only I think AWS um uh has Quinn had added some information in the previous Excel sheet, but I think he hasn't been joining this sig.

D

Right, um yeah I can, uh let me go over, what's been added and uh you know people that have contributed to the previous docs and stuff and I can start reaching out and see if they would like to do the um you know, uh information copy over here to this this new format or something and then we could go from there yeah okay. Does that make sense, yeah.

A

B

um I have one uh question to um you know in case you guys know so I think there is a new instrumentation and uh semantic convention stability working group that's being formed um now my understanding is the the semantic conventions uh that exists. As of now uh do not talk about. uh You know the context in which that attribute is required. They'll say a certain attribute. The requirement level is required, but is it required under you know, under which instrumentations uh in a span?

B

um What what? What? What do you think you know? Should we um ask for it? Will it be helpful.

D

Sorry I, don't think I'm uh quite following it out it's it's also the the vacation lag I. Think sorry about that. um Is that a question directed to me or.

B

um No in general, maybe you know others can answer too. Okay.

E

Yeah I think the Thematic equation work, that's being started which really started with ram being out so you've completely in the dark on that one um I think is really just about standardizing the existing conventions. I, don't think, there's any existing plans to say this must be included with this instrumentation I. Think it's just a case of these are what the standard like what a span is going to have. This is what this means.

E

I see that as actually part of what we're doing when we're defining an event which is really the outcome of an instrumentation from a client perspective anyway, that's what I would think we're doing is saying we want this event to be here, and this is what should be day um part of the rationale for putting it in event.data. So we just say your event, browser navigation.

E

This is what it looks like um was to effectively imply that as part of it, without having to go through and have a semantic convention for every single field which may be shared across different events.

B

But where do we capture that information um in in the form of a spec or some document.

E

um Yeah, so there was uh your original one, uh defining the API uh Bob, which I mentioned the event.to, and we decided to push that into later, so as part of defining the shape of an event.

E

um I think if we have some of these events that we've got in in Ram's document- and we say this is what it is. We start proposing um the definition and say the shape of an event. It looks like this here are the the four or five that we talked about last week um and their proposed fields, and then we get illicit feedback from that, except we still also need to do the the reverse mapping to say.

E

While we have a log event, this is how the same event could be represented in a span, and that could be where, like you're thinking here about semantic conventions for the URL in terms of and what you're talking about now is where the overlap is yeah.

A

E

Think once we start talking about attributes in spans, then we're back in semantic convention, territory, yeah.

B

I think there are two uh you know different, but related topics. I think one is introducing uh the standard that all the event data should be in an attribute called event.data right, and you know that uh can go in the API spec, and that also requires that uh the uh in the trace API supports attribute values that are maps right, so it has some dependency so, but that is one topic.

B

The other topic that I was talking about is that now now that we have different events listed in in the document, uh you know- let's say the page view the Ajax and the fetch, so each instrumentation will generate a separate span now each of them must contain these attributes. Optionally, you know, can contain you know these other attributes. So that fact, where do we capture.

E

uh I would think that's a side effect of the outcome of Ram's document. Where we're saying you know the Ajax event. Has these fields yeah.

B

But it should be consumable by like machines right. It's not. It cannot be only in a Word document.

E

uh Yeah, so so that that is where we end up going, but in terms of figuring out what those fields are to start with. That's why we're starting yeah.

B

Okay, yeah no I think that that is where I was wondering whether we should take up with the uh you know the other working group that Josh has started to see if we can come up with some mechanism to uh you know indicate in the ml files that you know this attribute.

E

um Reverse mapping, where is.

B

This universe, mapping, yeah yeah.

E

Where's this convention used yeah um yeah I'm, currently signed up for that. That group sorry um hazards kicked off yet yeah.

B

Yeah I know I I will be uh participating in that group too.

E

So yeah these are things we can cover, but I I don't think. Currently there is any idea of usage mapping, um but it probably would make sense, but it's going to be a huge task.

A

A

D

um I'm a bit lost with this reverse mapping thing I'm, just trying to gain a good night hit it up on that work. Yeah.

B

I can clarify um so today the attributes you know they are mentioned uh in a Solo form. They only say uh okay. This attribute, uh you know when present needs to take this ship um but which, uh let's say in a different instrumentation libraries uh create different spans. Now you know it, it does it. Let's say library, a includes this particular attribute, but Library a does not um so so the context in which that attribute must be present. uh You know that is not you know specified.

B

So, let's say in your document we have, let's say 10 different types of events.

A

B

We say that hey event, type a must have these three attributes, so even type B must have these. You know the three other attributes and let's say one of them overlaps.

B

Now that fact uh it is not possible to record today with the existing tooling.

D

E

um So yeah so the surprise, a little differently, the semantic intentions, Define field name and what it should contain. So they don't tell you where it should be included or when it should be included, which is the reverse mapping aspect.

A

C

But the current specters include the grime level versus if it's required or conditionally required.

B

Yeah, but it doesn't tell us uh like how do we validate that required um like? Can we build some tools to if we were to build some tools to verify the required requiredness? How do we go about it.

D

Are you talking about event, validation when the event is generated? You know yeah.

D

I think we touched on this topic before that is. uh Is it? Are we thinking you know? We need this uh at one time or at design time. No.

B

So there is also another reason um you know this came up in my mind. Yesterday, I think there is a uh we recently uh Martin recently introduced. uh You know browser uh namespace attributes to be recorded in a resource coming from browsers.

B

Now there is a class called browser detector which detects the browser and adds some resource attributes, so they have been using. uh You know a different set of attributes with the prefix process, dot, runtime and uh and now I feel, like you know, now that we have the browser, namespaced attributes, you don't need the process runtime.

B

So if we were to remove the process, runtime attributes assuming no one is using, uh is, is there an issue? uh Is it recorded anywhere? uh There's it's it. There's it's a loose banding today uh um and and that's where I'm wondering should that binding be more tighter going forward so that, if somebody chooses to you know, use the attributes in in uh in newer places, then that must be recorded somewhere.

A

D

I see so uh uh okay, I, I I, think I understand where we're going to. Maybe it means you know, we use concrete examples. It'll be much clearer, I. Think there's one thing that I've seen in across events is um a URL I mean, especially in browser right. Url shows up in every event that you are talking about. um You know. They're doing you know so semantic conventions is probably going to talk about. uh Url must be called search. The format is this: um it's required or not.

D

I think you you guys's question is that um is required? Does it mean URL is always required, or does it mean that when URL shows up inside of this particular type of event, it is required in that particular event, it's not required or something correct, also inventory of where, where all urls used. So if, if we're doing some, you know skipping I, don't know changes or whatever is that excuse me.

D

I also got a cold when I kind of came back on occasion, so policies, um so is that I I think I think I understand where you're going, but you know is how useful is this uh I'm just trying to look back, you know, I would like us to look back at our systems that we've built in the past uh internally right. Do we have such things?

D

um You know I'm, trying to understand uh the use case of that right. The usefulness in the use case of that it's it's it's required, but you know once we record it. How are we planning on using it? How do we intend to use it? I I'm I'm, breaking my brain trying to see if we have anything internal like that, we don't seem to that cross reference is reverse mapping or something. So the use case that you guys are talking about is still seems like uh I I, don't know this it doesn't.

D

You know you know really well, if you have practical usage for that, I would like to hear that.

B

um The Practical usage is it's more like uh you know, safeguards right, I think it's it's like. It avoids uh potential issues in in the future. I.

D

Understood: okay,.

B

Example, I just listed was uh there was a process.runtime.version uh that attribute was used so far to indicate the browser user agent right. uh You know, uh but now that more folks are looking at this more closely, we came up with new naming, and so how do we um deprecate the older convention? We cannot deprecate the entire attribute, uh because that attribute is valid in some other context correct.

B

So what we want to duplicate is its usage in this particular browser resource.

D

Understood, okay, um I I, it makes sense. Yes, uh the the flip set of this. um You know uh again, I'm not opposing this as I would like to you know, take a play. Difference indicate here, also the the useful, the usefulness of that I think we just talked about it uh yep. uh It could be useful, especially when you make breaking changes.

D

We would like to understand that all it's being used, so we can go touch them and uh you know make sure we don't break them and things the downside of that is keeping that up to date. You know unless it's it's automatic I think that's where you're going with some tooling or whatever. If it's not automated um the guarantee, you know we need if we start recording these things right, we need to guarantee that it'll be up to date and we have the right Tools in place that allows people to uh not just wear off.

D

You know if all we're seeing is we're just going to document somewhere and uh the manual process is such that when you anytime somebody starts using that in their own usage, they have to come and record it here. It seems like a really weak link right. It could break yeah.

E

Yeah and that's all there is today in the conventions like even in the the specification um it's a manual table to Define, which language supports, which part of the specification, whether it's optional or not, um but yeah yeah part of the the issue is unique to where things have been coming from, because it's in the past it's a it's been like a hack to say, okay.

E

Well, we don't have anywhere to send this field so we'll send it as a process runtime value, but now we're stabilizing it's um it's it's going to get mapped, uh it's probably equivalent to like internally. We have like all the EXP extensions, so we have the one source of truth of where to go. That doesn't really happen today in open Telemetry. It's just the case of we have a list of attributes and when you've got like you want to represent a HTTP connection. They say: okay, you've got to have these resources.

E

uh These attributes, with these names and oh by the way, this one's required so.

E

um But yeah there is nothing for an event today, which is why span events have the half they have today and we have to Define it for what we're doing foreign.

E

Effect of what we're doing will produce the table to say event: X attributes your accurate map y, um but that doesn't preclude someone going according instrument, an instrumentation and using some other attribute.

E

uh We can't control all of those. So all we can do is say we we Define what the open Telemetry events are going to be and what should be validated. Let me move forward on that.

E

um I guess, Santos to more directly look at what you're talking about and probably for T2 as well, it's a case of if we want to deprecate the usage of process runtime.

E

um What I've seen in the past is effectively we publish book um at least an open Telemetry world, so you publish both up to a certain version and then the next version drops the older one. Okay, I think that's what tignan was proposing for um a bunch of collective changes.

A

Yeah yeah, we can do that yeah.

B

B

Do you guys use uh that browser detector uh class.

C

No, we just take the browser attribution Center, not our own right for uh user agent. We do with the ninja ninja site. So we take the request, error, yeah.

B

From the OTL per request,.

C

A

B

um There are a couple other topics, actually these this agenda- I put it for uh tomorrow, but we could discuss about today and tomorrow. Okay,.

E

I'll say today back to tomorrow, no.

B

B

Foreign I think before you joined, uh we were talking about the the first topic, which is uh we want to include the URL of the web page. You know.

A

Which is the referral yeah yeah.

E

I've just been looking at your things so that that would be the referrer um like today, at least for our internal stuff. We pull that out of the uh the header so um but like the same way that the user agent, we scrape the user agent, and they refer from that. um uh In fact, he did not I think I included a separate field in the document went out there actually.

B

You you uh did, you include uh um the event for Ajax.

A

B

A

Yeah I'm, just looking I, think I might have put that in there.

E

E

I'm pretty sure we do have an optional field for that.

B

Okay, so yeah I think what we call is is one thing. um The other thing is you know: uh how do we send it to be in? We earlier at one point said we'll send it as a resource level attribute that can change, uh but now that the research level, attributes I, think that's. uh You know stuck uh I'm I'm wondering if we should get going with the scope level, instrumentation scope, uh level attribute.

B

um It will at least be once um for a group of uh spans generated by that Library. So so, if on a web page, if you make three uh xhr calls and two fetch calls, then you will get this attribute once for each group, yeah, which is still some saving.

E

But yeah it's like the the referring, so even in a spa you would go and change that um yeah I did document it there. For some reason, I'm gonna find it.

A

It's one to too many windows.

A

It's probably gonna take a while, so we can talk about something. That's why I'm searching.

C

uh Anyways, that's uh too long didn't read them a comment. One thing I did think was important that this is the current page. Url is also more important for, like non-xhr.

B

C

Yeah for like user interaction or when you're instrumenting, graphql or.

B

C

Anything in the browser.

B

Yeah, except for the base page uh you'll, need for everything else. Yeah.

C

So I I don't know if I don't know if, if like refer, is the best for non-hp use cases.

B

Yeah yeah I think yeah. We we should change that. I agree. Okay,.

B

Yeah thanks for pointing that out.

A

B

I didn't read carefully yet, but um you also mentioned that the base class, if you were to put it, um put that as part of the instrumentation blessed class um there should we do that, given that that class is used in other places too. So is there a base class instrumentation based class specific to browsers.

C

Currently, thank you guys. The node and browser interpretation basis. Major classes are completely different. For example, one has a double enable protection, other one doesn't which one doesn't uh but I would rather not have such differences anymore, if possible.

E

Yeah too much of a class hierarchy, it causes the web to break into much code.

B

um Sorry, I didn't uh totally follow it. What what is the difference today.

C

One is a lot one, there's a lot more features than the other one. Basically, I'll find the links.

A

Yeah I'm, looking for where.

E

The Define I didn't think there was a separate web instrumentation base. I know, there's an instrumentation phase.

C

Is common as long as there's a platform folder there's a massive incoming.

A

Oh okay, what do we got any browser? Okay, the expectation.

E

Okay, okay, yeah yeah. It's really basic, though it's just yeah yeah,.

B

Actually, um if you see my screen, um there is a yep there is this instrumentation base, but it doesn't seem to be used. um So there is a same.

C

Instrumentation.

B

E

Everything everything will extend from instrumentation base.

B

Yeah, but there are two instrumentation based classes, one under browser, one under node, yeah.

E

So so, when you're.

B

E

Application yeah it'll um when you package it it will change which face gets used. So if you're, if you're packaging for a browser, it'll use the browser one, if you're practical packaging for node will use the node one I see.

C

A

Just an awful way to do things.

B

I see I I think the reason I I think I got confused. Is this uh index.js at the platform level it? It includes this.

E

Is no yeah and.

B

Also, let's say if you look at the fetch, instrumentation yeah.

B

It uses uh the the templating uh argument, which is not present for the browsers yeah.

E

In fact, looking at the node one, the node has public methods is enabled disabled, but disabled and enabled, but the web one only has uh or sorry the browser one only has enable, which means, if anything, uses those other methods.

E

It's going to like go kaboom.

C

Those methods are still implemented in instrumentation, abstract version. Okay,.

A

In that case, it will cover it. Yeah, yeah.

C

If you compare the node and the browser versions, then you can see that the node version the disabled is extended to also another check for, like the AC is already enabled check, enabled force on patch, while browser one doesn't have anything yeah.

B

um What what about this I think? Where is this um template argument coming from.

E

No, almost a generic, so that's.

B

Going to be yeah so that generic uh is only I see it only in the nodes version of instrumentation base, not in the browser version.

C

Typescript only checks, typescript is ignorant to browser field, so it only uses the node one yeah.

E

That includes the one that's defined, gonna, give a whole bunch of other side effects.

B

Okay, foreign, but yeah I think going back. We we can and I think we should figure out a way to include those uh that browser page URL um as a scope attribute for for all the functionality.

E

Yeah I did a quick. We only have um her page view and Page actions, so you know clicks or user interaction, I think before the dock we don't seem to track, prefer URI for uh the Ajax requests, which means we're gluing them together, based on the session ID in the back end today, but yeah I agree. It would make sense to include it.

B

Okay, um what do you want to call it now.

E

um Referrer is probably the what I would go for so.

B

Referrer um I think it makes sense for xhr and fetch calls, but let's say we also want to use it for um user interaction. Events.

E

um We already have that in Microsoft today, and it was we call it referral, that's the pages coming from.

E

We actually also passed back in the referral, as well as the current page name and values. Do you remember offhand Ram, which.

D

No I need to look up um yeah.

A

B

We could we could introduce a new name saying browser.document.url.

E

Or just page URL yeah.

B

Browser.Page dot, URL yeah.

E

Because yeah we have uh request URI, which would be the URL. We have page name, which would be the document title and we have page type which I don't remember. Why that one was there.

D

Yeah, the page type I think is just an.

A

D

Thing um yeah, if shopping, cart, page check out page that that bit of stuff, if I remember correctly, that's.

E

A more Behavior stuff, yeah yeah.

D

So uh remind me again uh so we're we're trying to you know, look for a suitable name for referral URI. You know, uh uh but.

B

Then no, we are trying to uh not necessarily a referral, uh a referral URL, but we are trying to capture what is the page on which these events happened. Right.

D

uh I, don't believe, that's the referral URI.

E

That's very good question you're right for page actions, yeah.

D

E

I got confused because we also also send back the referrer if it's present.

E

A

A

Yeah I think we can settle.

B

With that browser.page.url.

C

B

uh Okay, I have one other uh topic um quickly. um I think this is I. I was thinking about uh now what you mentioned uh last week that uh like currently for the document load, we are generating spans and- and you don't want to wait, uh uh you know too long. If the user chooses to navigate away, you know, so you want to send an event for the page load right away.

B

um So I was thinking that we can. We can still do that as uh on top of the existing spans, um but I uh have some questions on. You know how that will work.

B

um So there are the two uh things you know. One is I think you will have to, since you mentioned that uh we'll have to include the trace ID in that event, uh you'll have to start the trace. First, you you don't need to end it. You don't need to even populate much details in the tray in that uh in that span that that that's part of the trace, but you will still have to start a trace so that you get the trace ID that you can put in the that.

E

You can include and Link them back together, yeah.

B

Yeah yeah, so so that will be one change, maybe that's doable.

B

um The second thing was um so today uh at least we uh at our company, we plan to use the batch exporter batch span processor to export the spans in in a batch, every 10 seconds or whatever seconds right now, uh even if we generate an event that event is, is going to emit from the logger right away, but it's going to sit in the in the batch exporter, uh so you might still have the problem so.

E

um Yeah, there's a bunch of stuff that we need to do to make the existing exporters browser ready, and that is um in fact there's one of them. I look at I, don't remember which one that already does listen to the unload events and then tries to flush, but I think it actually tears down rather than sorry shuts down, rather than um flushes stay open, yeah.

B

I think there are uh complications on that part too. uh One other gets.

E

Back to the bachelor, the the event size that I was pushing through day, one yeah.

C

uh It is the current patches spine processor that does force flash on visibility and pages.

B

But yeah in this case, uh can we Mark certain attributes in some way so that they are high priority events? And you know they don't need to. They shouldn't wait. Yep.

D

I was I was going to suggest that I think that's that's required for browsers.

B

um Our one more option could be that you, you use a completely different pipeline which doesn't use uh a batch exporter.

D

For high priority events right.

B

For high priority events, yes,.

E

Yeah so internally you can have two batch um ones. One.

A

For higher ones, for normal.

D

And that's what that's! What we have internally in our uh yeah I.

E

Well, it's sort of all unified, so it's actually one process that understands the the different levels: uh I'm just gonna, open, Telemetry perspective: I, don't want to introduce another class just for that, so what we could manage it with two instances today.

B

uh But then there is a runtime hit right. I think you know we will end up uh creating uh a parallel set of objects in memory for the second pipeline.

E

um Well, a parallel set of, like your events, uh sorry event, um timers and stuff to be able to know when to fire it and stuff like that.

B

A

Right I think starting.

B

From the Tracer um to the.

A

E

Where you put it so we could yeah because.

B

The exporter has to be configured in the Tracer uh and therefore, if you were to use different exporters for these two pipelines, the Tracer will have to be separate too yeah.

E

Yeah yeah, it probably would be nice just to extend the exist.

E

You want to understand um well, probably probably understand a couple of different things is you'll have higher priority events that you want to send as soon as possible, but not currently normal ones, which you can just go on on the default, and then you will have some super high ones that you want to like send thicker, so you want to send it off now and the current um uh you know JavaScript execution cycle, not in the next one, the advantage of the the critical one and doing on the next one is.

E

You can batch more more of them together where secret, if you say, send it now you're only ever going to have one event. You know in a request: um yeah I think we. We should probably add that to the list. Okay,.

B

So I'll create an issue um putting these thoughts in it. um If, if we can do it with the with the existing uh infrastructure, I think you know there is a runtime hit yeah, um but if we were to introduce a concept of a priority, uh you know that can escape the batching process. Foreign.

E

Not necessarily.

E

um Because it's not it's, it's very client-centric.

B

um Yeah but the behavior of the patch processor will change.

C

If you know it comes from one Library, if you use batch processing, whatever it's done, but uh I'm scrolling back a few minutes, you can register multiple spanning processors on one laser provider.

A

B

Is it yep, but then both will it will go through both right.

E

Yep yep yeah one needs to be able to drop the other yeah.

B

um So you'll need to add some filter.

C

Or you can just have one span processor and put an if in there yeah you.

A

C

Always do anything you want. Okay,.

A

E

Specific batch processor, as well like, which is why I said it, doesn't necessarily have to change the spec because we say for the web. We do this and that's how it works like today. The node won't be listing download events because it doesn't exist, um I think when I also looked at the batch, the processor, it doesn't actually list to all the unload events. So that's also another change that has to happen. I think it was only listening to unload and before unload I, don't think it listens to visibility, change or page height. Today,.

B

So here is another reason why it'll be good and good to be in the spec right I think if we can better than it's it's it I know it's making it complex. But how do you um indicate which attributes are high priority attributes.

E

Well, it's not attributes it's the event.

A

B

A

It's the event: okay, yeah, yeah, I'm, sorry,.

E

Yeah and and again you're, coming back to to our working group, we can actually say as an event we Define the shape of it and we could say as part of defining an event. We could have a a priority field, that sort of says your normal critical.

E

um You know critical or you know real time whatever you want to call it um and then, as part of the uh batch processing, we can say well anything, that's processing this event, whether it be a span event or a span event is going to be fun because it won't get sent until the standards, but anyway uh whoever's processing. It can then say. Well, you know you should look at this. This attribute of the event- and you know deal with it as such.

E

Okay um and the reason I keep pushing back on the spec is a case of this is very specific to the web. Only so it won't apply to pretty much it or not or sorry, clients, only not not the web, specifically but clients in general.

E

Okay, so I try to get a spec change for a back-end server. That has no concept, no need for this. I think that would be harder.

A

B

Yeah I think if we are talking at the event level, uh yeah I missed that yeah okay makes sense. Okay, um I'll I'll create an issue and put these thoughts. There.

E

Yeah but yeah the the web web ready version of the batch processor is a bit like that stupid instrumentation, based extension, that that just needs to change um so again. This is something that we could do in the um when we get this.

E

If CLA unlocked in the sandbox to go and play with this implemented, validate that it's working and then push the changes back, I'd like to remove a lot of the the node browser, Forks that are currently there because they're just it complicates the crap out of it and it reaches typing issues like you can on Santosh, where node has the the typing, but the browser one doesn't and hey Presto you're going to have pain at some some level. If you try and perform too much.

A

Okay, uh that's all I had.

B

Suram uh you you uh you'll join next week as well right I! Think we could uh discuss further or are we are meeting again tomorrow, uh but I don't know if we'll have enough time to to review this.

D

Yeah uh next week sounds like a plan: okay, okay,.

B

Anything else what they do.

A

A

Okay, all right uh I can't hear thanks. Everyone.

D

A

A