Open Telemetry Uncategorized, 5 Apr 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2023-04-05 meeting

Description

Open Telemetry Meeting 1's Personal Meeting Room

A

Hey hi everyone. Thank you for joining all.

B

C

B

Me, oh hi, everyone.

A

Okay, I see Karthik here, hey hi, Karthik.

A

Yo, let me share my screen. Probably I think we can start discussing this.

A

Okay, uh yeah I think Karthik. Probably you wanted to talk about some some issues that you are facing right uh in open, Telemetry, C, plus plus usage. Normally we don't use this forum for discussing any troubleshooting or discussing anything any issues, but I. Think probably uh you wanted to talk about something I. Think I just told you that you enjoying the meeting. Yes.

D

A

So probably I think if it's fine, we can wait, we can wait before I mean we have some something in the agenda. I think we can discuss that and then probably we can come to you sure.

D

uh We can take up my question. Queries at the end.

A

Yeah thanks thank.

D

A

So much okay, yes for the specs, I! Think Mark. You have written items, specs release 1.20 in preparation. We need to adjust the semantic convention. Yes, yeah I. Think probably we should wait for a new releases.

A

And we don't have any issues right in the semantic conventions. I mean the source generation, so it should work seamlessly.

B

Yeah resource generation is working now, so I expect it to be cooking.

A

Okay, I had added cup I. Think three items from the specification. One was the supporting the elastic schema in open, telemetry I mean this is right now part of Otep, so I don't think this is yet finalized or it's already merged, so I think it's already finalized. So basically there would be a consolidation of the common schema schematic conventions. I mean the schematic Dimensions uh from elastic common schema and the open telemetry, which means there would be further changes in these similar current existing somatic convention and I.

A

Think I was just going through the ECS uh schematic convention. This looks like most of I mean there are some deviations from the way they they support, HTTP based conventions. So probably there would be some changes, I think coming in the semantic convention, so we just need to watch out for this, um but that that's a good thing in general for open Telemetry adoption that if it's more of a consolidation of the conventions used across the industry and if everybody is going to streamline to use the open television for that.

A

uh Hey Josh I think you are you in the correct meeting right. It's the simplest Place meeting.

E

Actually hi, yes, I, haven't come to this meeting in Forever um and I did show up just to like uh drop in and chat if I, if I may, I I don't know. What's what's on your agenda, I thought I'd, listen until you in an appropriate time.

A

Okay, uh sure I mean you have anything in the agenda or you just want to.

E

Listen I was about to I, was about to open it. I can see what you're, discussing and I um I wanted to raise. An issue about Envoy um is what I wanted to talk about, and I and I was hoping to to come and maybe ask some questions about Envoy. Basically,.

A

uh Okay, I mean probably if you, if this is something probably you can discuss right now, I think. Probably we can do it and then you want feel free to drop it.

E

Oh sure, well, that would be kite um of you. I um I um came to Sacramento as I mentioned and I'm. Putting this in the chat and I can also open your notes, but I didn't have them open. Yet um here it is um so the the history that I'm coming in with is not to pressure you in any particular direction, but there's a history that I have here at lightstep involving a lightstep tracer which came from open tracing and had we had a contractor working on that forever.

E

um Who did a really a ton of really high performance uh work on that Tracer and lightstep? Had the C plus plus it's C, plus plus Tracer as part of envoy from the early days it was a privilege of working with Lyft when Envoy was pretty new um and so Lyft put the light step Tracer into Envoy, um and until there was an open, Telemetry Tracer available.

E

We we kept that position and then what what happened was Envoy noticed that light steps Tracer had was very unmaintained and and became looking at us to do something about it and I said no way. We're going to maintain that anymore, which was accurate and honest, because I know how our company works, but the the push, but, but actually what happened as a result was online removal concept Tracer from its main release and now we're telling customers to use the hotel Tracer, which is fine and I, noticed some performance issues that I wanted to see.

E

If that had ever come up here, and this may not even be an open Telemetry issue as much as it is an envoid issue, it looks like there's no way to configure the open, Telemetry, C plus plus Tracer, when it's running inside of envoy, particularly for performance, and one of my Engineers had pointed to a line of code that does the configuration of the hotel, Tracer and there's like a batch size parameter.

E

That's five and and and we're dropping spans and that's all I came to talk about, um was to see if anyone here is aware of the integration with Envoy or whether someone has here been here discussing performance and the integration with Envoy dropping of spans or high throughput in any sense, is what I'm here to talk about.

A

No I think we haven't heard anything about Ry at least integration with NY. So we don't have any idea about that. But.

E

Okay, that's fair. Can I ask a couple just sort of like the most basic question that comes to mind is like what is and I could surely. I can go answer this by reading the tread myself um so I'm. Sorry to do this, but but something like the traditional um organization of a tracer has been like. With a span.

E

Processor has some Q size and some timeout parameter, and then the exporter just stands and there's basically one export at a time and the problem that we've always seen going back years and years with even all of our tracers as well, is like you have that design the buffer fills and you flush and then, if, if the, if you can't flush fast enough, the buffer will will start to overflow.

E

So you then you say: okay, the Buffer's, not big enough, so I'll make the buffer bigger, and that makes the request bigger, which makes the request slower so that no matter what you do, having a single sender will just never keep up batch sizes. Don't work, timeouts, don't work, um and so that's why this light step Tracer had been engineered to have multiple concurrent senders and to do some fancy stuff and I.

E

Don't even want to talk about the fancy stuff, but I just know that having an ability to um create more than one sender is usually the solution, but most Hotel sdks haven't done that at this time and the result is Envoy can't can't be used for, say, 100 tracing, which is people. We know people that want to do that, and so I was I was here to talk about it. I want to find now the piece of code where that happens, but and talk about it just for a second.

E

um Let me go find that link um sorry to make you wait. I I know.

E

um Where was that um and.

A

Which is which is exported they're using I mean well.

E

This this would be the otlp exporter.

A

um In this case,.

E

Okay um and Where is the.

F

C

E

So there's there's something about Envoy configuration and you, as you can see, I should have come a little bit more prepared.

E

um So let's see, there's a batch size parameter. I'm, fine I! Have it now, um okay! So let me put in a link now to it's a premier link to some GitHub code in on boy yep.

E

um So that link um just shared is a place that we think inside of envoy that the Tracer is configured and it has this Min flush spans parameter. That has the appearance of something that can be configured and we haven't figured how to do that via Envoy configuration, but it.

E

And maybe this is this is not even the same well guys. I. Think I've made a mistake in coming too soon to this group.

E

um I was hopefully going to hear that at least somebody had talked to you about Envoy, and the answer is maybe not um what I plan on doing then is to research this a little bit further.

E

um The the end result that I can say is that when a customer- or we internally run our envoys with tracing at 100, we drop a lot of spans and um we would be interested I, mean I, say we're interested, but obviously we're not putting forward effort. You know that old lifestep tracer was performing well and we're we're still running the three versions of the old older Envoy because of it um I do wonder if there's anything of value there that could be. You know, sort of unvolted and put into place to do this.

E

um uh Basically, it was a zero copy, lock, lock, free algorithm that was essentially building request buffers as efficiently as possible and then sending them in a it was. It was really complicated.

E

That's the problem I'm afraid of, but it did perform very well um and yeah, whatever whatever it looks like this, this code, with a five value that was hard-coded unless you have a configuration that we can't figure out a set that seems the most suspicious to us and we were and that's what set me off coming to this group and just chatting right now.

A

Okay, yeah difficult to difficult to understand uh what exact exactly is a performance issue by looking into the code here, but I.

C

Think it would.

A

Have somebody from one by definitely chipping in in the meeting and probably you can explain really what exactly.

E

They're, seeing.

A

The issue I mean.

E

Then I I will conclude my topic. um I will come back with more background or more people or something but I've at least spoken. You just Express the concern and if you heard about Envoy, you tell them come talk to me. Thank you. um Yeah I hate to just drop in and drop out, but I. Think I will now leave your meeting and apologize for interrupting it. Thanks.

A

Thanks for setting the context, bye.

A

Okay, yeah, that's that's interesting, I mean at least haven't heard of any performance issues, still law, uh but it definitely it depends on the I mean to me, it depends on the throughput of the traces number of traces or number of spans which are getting created. I mean ultimately.

B

A

Would be a limitations so.

B

Just some some comments on that uh I think you mentioned to use a piece of a batch span processor uh in respect. We are supposed to be environment, liable to size, the king of the bsp and I. Don't think we support those. We support sitting problematically the sizeable but I, don't think we use the alignment.

A

So that's what I felt probably was it was it was it it's talking about setting the environment very well and getting the and updating the queue size at the runtime. Was that the was that the concern foreign.

A

B

Yeah, so this thing is not using environment.

A

Yeah, so we don't use that so the two sizes are at least uh defined at the these boot stopping time when the when, when the SDK starts and there's no way to change it at the runtime. But.

B

Or you need to to write code to change that, depending on the basically.

A

B

A

Think there is anything.

B

A

The specification.

B

A

For the environment very well probably would be interesting to see.

A

I can't I can't see anything here.

A

Okay but yeah, let them probably come up with more details.

B

Okay and another comments so I, don't he said, don't kill people don't know if it's your PC or http1, uh who went with a synchronous, export uh I think with a synchronous, I think uh multiple, the pixels.

A

B

Well, I trash myself, so it's mostly I came in I. Think um one thing uh with asynchronous export I think we can have better currency and better for good. Also.

A

Yes, yeah, probably that was something we wanted to talk about, but I'm not very sure whether it's totally specs compliant or not, uh so that that's the reason why I'm not explicitly talking about the async support at the grpc level, we do support it, but probably that's something. Yeah.

A

I mean that that was only concerned whether it's it's uh I aligned with the specification or not because as per the specification, uh what it says is if I'm not wrong or if it has not changed till now, then it's used to say that the processor should return back when all these spans are exported or or the export method should return. When all these paths have been exported and I, don't think that's happening in async uh if we enable the async plug so okay yeah, that was that was one one of the things but yeah.

A

Let them come up and then probably we can provide more options.

A

Yeah I think in the interest of time we're probably uh there are a few more tips which I think talks about supporting is remote flag in the otlp I think it's still the Otep. So probably getting just wait for that and again the infamous ad link support, which has been being discussion for long I. Think still it is not finalized. So there have been talks of not supporting the adding The Links at the at uh I mean of the Spanish created but supported using, add event, but yeah, that's ongoing things.

A

So probably I think you can still wait for this yeah, uh so I think that's that's! For the specs and I think we can go with the miscellaneous I see. There are some important things to discuss in the miscellaneous and then you can come back to issues in the pr, um so PR review clarify exact number of approvals needed for the pr yeah. That's that's uh was something on me or to basically Define I mean I was planning to read to raise a PR uh to get the suggestions on.

A

How should we do, or should we handle this in terms of number of approvals based on the size of the PRS I.

F

Think we have such a raw right or that way.

A

Yes, so it's not defined based on the size of the pr first of all, and we so probably if we can make it make it more explicit and.

A

So we we just say that if you, if, if it has received two approvals with at least one approval from when a brewer and maintainer at different companies, uh so uh it means that it need at least need to approval, irrespective of the size of the pr which is very more opportunity. Okay,.

C

A

Probably we can.

A

I, don't think we need, for at least for the Cosmetic small documentation. Changes to approvals are required. So probably that would be good good to have more clarification in terms of the kind of PR. It is a pull request opened by the approval or maintenance, only with only one approval from a program which makes sense, yeah, yeah major feedback items points are resolved. It has been open for review, for it is a day one day, one working day, I'm not sure again.

A

This should be depending on the size of the pr for a big Freer like HTTP, SSL I. Don't think one day is enough, um so probably we have to do some changes here.

A

So I think this need just needs. This need a rework. Probably we have to.

F

I see yeah I, agree yeah, so I mentioned. Maybe we need to mention the the size of the pr may affect and.

A

F

Small PR can be expected right and then, but you still relies on the maintainer to apply or program to apply the correct label.

A

A

I think it could be done uh probably by Yeah by one of the approvers or even the maintainers can do it yeah and then yeah for any request which is created Max. You want to add something on this. uh If you have.

B

Not ready, it was just to clarify, especially for the assistant yeah.

A

Sure I think probably let let me raise a raise a PR with the changes. I know that projection point on me: I'll. Do that.

C

A

Week and probably let's, let's all comment on that and if we all agree I think you can Define the more explicitors for that.

A

You know what this one close: the Legacy projects, 1.4.1 release and Matrix SDK implementation. What is this.

B

Well, when you go to projects, there is a new one, but at least I'm creating the Bell to two other things.

A

Yeah I think that makes sense.

A

Or we don't see it here right, you have to go to uh still have to get too used to using projects I'm still using.

B

The Milestone so go to projects over here.

A

B

A

Yeah, okay, yeah and.

B

Classic: okay, yeah.

A

This this is created long back, so probably I think I think I can close. It.

A

Is there a way to close the projects into.

B

Well, no need to do it now that I was just.

A

A

Mr holiday, yeah I think that's even I won't be available on 10 10 11 I'll be on vacation for these two days, 10th and 11th. So.

C

A

Tom, what do you think should we uh you want to have the meeting I mean I? Can we can leave it to you, I mean you and as on, uh if in case me and Mark is not there.

B

um I think Houston is on drama in Germany, so uh I'm not familiar with the Eastover about time. I would be assuming it's also hungry.

A

Okay, yeah make sense yeah.

F

So, okay, it seems I've seen since we we need to cancel the next meeting next Monday.

B

F

F

Yeah, it is question to me and yeah and as we can always discuss in a slack Channel and there's no blocking item for for me for that here,.

A

You have planned the release for 1.9.0.

A

I think we didn't have a release for some time right now um and you can quickly see yeah. We have another meeting last month so and.

A

I think that makes sense I mean you can have one. What is it should we wait for the specification to have a release and then do a release so.

B

uh The minimum I want to wait for to merge versus encode, uh which I will be doing.

B

A

Yeah I, don't think anything new is there. As of now I mean you open PR, there is one PR open which probably I have to review it and I'm not getting time for this for this, um but you probably we can even try to you if the PRS I think there's no there's no comment or no no issues in that. Probably you can even use this also, and this also.

A

Okay, uh yeah talk: uh you want to talk about the user facing blogging API.

F

Yeah yeah sure I think yeah um I have this um thought because I think currently our login API in our repo here and keeps in preview for for a very long time, maybe two or three years, and then we we even our API, we changed it yeah we did some major change and maybe one or two major changes for API, so yeah. One issue of this is this um makes makes their adoption or instrumentation a little hard. We.

F

Maybe it is hard to persuade the customer to instrument and we can't guarantee sometime later you need to, but we'll re-instrument all of your code with on our login.

C

F

Recently, in this back in the spec meeting, the discussion or in a log um logs back meeting interest discussion about about decouple the in decouple, the bridge API and the user facing API right so so from what I I get is that the the old test Bank were not enforcing any user facing API. So so the language itself needs to either create or adjusting bridge to the existing login API and based on the bridge position API and while the original API, the naming, is still another discussion, whether it's just a coin Bridge or remove bridge.

F

But I think it is for sure that the language can have its own login API, like for maybe not for C, plus plus for other languages, there's well known on logging Library. So they are not going to change for for hotel.

F

So even in future, the current the current Bridge API change, because the log spec is not stable, I think the the user-facing API for the language were not changing. Maybe they just changed their appender or implementation to say to call the new bridge API, so so I think that's a current status so for C plus I'm thinking about.

F

Maybe we should do create our own login API I mean create, or maybe an appropriate word is just make our some minor minor tune into our current area, and that means we create from scratch and discuss in this card or current apis I think the current API is already very common, but maybe need some minor adjustment, maybe yeah for for most scenarios.

F

I think one reason is because I think currently the API is I, think I only hear um Owen and he's using it and I have be I already checked with him that he he's using it like, maybe nothing. You know large volume, not in not high throughput, maybe just for a few, even a few logs, but maybe then a few locks. So the the condition may change if we like to want to um want to instrument for some like multiple components like that and then yeah so like the unstructural log will be inconvenient.

F

So maybe, as I want to work on this, to make our API complete and let's see yeah in either even in future, there are spec changes or some somehow the Rich API currently required changes. Maybe we just need to change our SDK implementation, like called different Bridge API, but are using user facing API is stable, so we, this is what I'm trying to do here so with this I think it will help us to say to help them to us and for adoption.

F

I mean broader adoption yeah, because the I think upgrade SDK is is um is expected for the for the for the libraries for the users, but to rewrite the code, maybe maybe not, and they don't expect that yeah.

A

Yeah I think probably just I agree with you, so for C, plus plus we don't have any user facing logging API. So the current implementation is for bridge API, which.

F

A

As per the specs, whatever we have implemented but specs.

F

A

Does not define any user logarithmic user.

D

A

Logins API, so we do need to come up with something of our own yeah.

F

I think this is um for what I think this is allowed, but speak right tendency spec Dynasty later I will create like the long swag from what I I'm currently aware. This is saying, maybe in some time later they were created their own API and then require each language to implement they. Don't um that's not the current plan right yeah.

A

F

They don't enforce.

A

A

So basically, the the bridge API so I mean the specs is not going to Define I, don't think at least in the in the short term, or even in the near future. They are going to define the user facing logging API. They expect all these six to have their own implementation either. It should be the the logging framework which is used by the, which is provided by the language or.

F

A

Not like for I think C plus plus, we don't have any. You know a lot of time, Frameworks uh uh specifically coming up as a package from a given like centered library or something so probably.

F

A

To come up with our own yeah.

F

A

So so only thing I would add it that that it would be good to comment out uh that whatever is the existing API, whatever methods we have in the existing API that are recommended to use as a bridge and then whatever we.

F

A

F

F

We comment out of the bridge API.

A

Not not government writer I mean the header right.

F

A

Comment that this is recommended to be used.

F

A

As a bridge, API yeah yeah.

F

I think this is a good thing: yeah yeah.

A

And then then, whatever newbie ads, because we have I mean we have to use the same space Bridge.

B

Api and logging.

A

Api, so that's.

B

F

A

Comment we can at least say that this is something to be used as a bridge API yeah, then probably we can with that. We can even say uh to make our user facing API as uh I mean stable, even though the bridge API is not stable, currently or still experimental. You can independently uh set the status for both of them as as.

F

Okay, yeah I think you have to add a comment. Maybe not put all the rich API into one group in the same class. It's fine but I think because uh I think the spec spec has not been still not very clear like for the future. Maybe they, the bridge API, might be also exposed to the user right.

F

The user can call or not we don't know so, maybe if country we just clip it there, but um for our user facing part I think that that one will be different and there will not be changed by this vehicle.

F

A

Yeah yeah, so so I mean that's what I was saying. If, if you plan to make user facing API as a stable.

D

A

Is still experimental, we have to do some somewhere. There should be demarcation that this is a big API, and this is a user.

F

A

F

A

I mean that that with debarkation I mean that you, even though we can add it as.

A

It as a in our code, just just do the header comment and say that this is this is something.

F

A

To use as a brick API.

F

Comment is one way: I, don't know why I.

A

Don't know yeah I.

F

A

Other way of changing it apart from comment, actually so that's what I was thinking I see.

F

Yeah, the part and the thing I don't know how to do is like um add some message to them like Market, something like Market has deprecated. We don't mark it as deprecated to just associate to prove this is not preview. You I think we don't want to do that, because spec is maybe some later it becomes.

F

It becomes fine to expose user in a bridge API, so just keep it there, but yeah and with comment I. Think that looks good to.

A

Me, though, I mean the only reason I'm saying comment is that this way you can mark your user facing API is stable yeah. If we all agree that there is not. This looks good simple enough, even though the existing Bridge API is still experimental. We can do that yeah. If we have.

F

So in one source file, maybe infos API in one header file in one class, part of the API work group and then the bridge API. We group them together and the same this. These methods are experimental, still and and a bridge API, which follows the spec older spec. But for the other group we see, this is our.

F

um This is for space for our Auto, C, plus plus, and we see we can declare that stable right. This is mechanism.

D

F

Signature or Methodist will not change or I mean what we currently. We will maintain background compatibility, I think yeah. This.

A

F

A

Okay, okay, so probably, if I, if I understand, you are working on this right user facing login.

F

Yeah, if you agreed by by all them antennas and the market for approvers I, think if we agree, this is a right thing. We we want. We need to do I, think I will work on this or Mark and yeah.

C

Sounds good to me.

F

Okay, okay, yeah.

C

Thanks I think for.

A

Yeah, for this part,.

F

A

Yeah, we look forward to uh any changes from you and thanks for taking this off, I think this was something we don't need. It's very important.

F

A

F

Thanks thanks, Mark and uh and at the first step, instead of just changing the API I, think I'm going to create a dock in our docs folder and to to list the ideas like the what change we want to make to existing logging, API and yeah yeah. We can see whether that's a thing we want to make that stable.

A

Yeah, that's a good idea.

A

F

A

Lot of time, probably uh okay- this is 15 minutes just coming back to you quickly. If you want to talk about the issue which you're facing I mean we do have some more things to discuss, but I think probably we can use five minutes to discuss and then go back to our stuff, I'm. Sorry to keep you waiting.

D

No, no, it's fine, I can wait, I mean if you want to take the discussion at the end.

A

No I think probably we will.

D

A

No, nothing go forward. We just have 20 minutes left, so probably I think you may not get time in the end.

D

Sure let me present my screen.

A

A

D

Yeah yeah is my screen visible.

A

We don't see anything shared, oh yeah, I think we can see here. Sorry, we can see here.

A

C

F

See your screenshot as well.

A

Yeah, you can see.

D

D

My requirement is very much similar to this city.

D

Client example: okay um I want to start a span and then I'll be making guys. I will end this one. Okay response, I will be getting, will be in a different uh function. It won't be in the same scope and instead of using HTTP client I'll, be using uh rqt's Network stack instead of curl I'll, be using qt's network stuff. So I think we we have.

D

uh You have helped me out with this in the group chat, so I'm able to send the basis I mean I can see it in the New Relic uh being reported, but current issue is in the distributed.

D

So what I see is when I Mark and active okay um I see that the trace parent header is getting created, because that is the current active span?

D

But but if I don't Mark that span as active, then I don't see that transparent header in generated. So basically I want to send that header to our backend request.

A

D

So my my question is uh without using this active span and uh uh without using this active span, how do I generate or send the trace parent data to our backend.

A

So you have to express explicitly create a context: object set your span contacts in that object and then call this HTTP text map carrier on that context object. So so, instead of this current context, which you are getting from the active span, you have to create this. Instead of doing this get current, the current context is going to get current. uh You have to create this context, object instead and then populate it with this pan context and then pass it to the text map propagator.

A

C

You I think there's line number 36 and.

A

30 song will change and let me see if I point I think there would. There should be some somewhere in the at least in the comments or somewhere. In some issues we may have give suggest. Given the suggestion, let me figure out and I will send you the link for that, so so the what what you're talking about is not using the current active context, because you can't use it because you are creating this pan in at uh at one place and then you're trying to probably populate it in some other function right.

A

Something like if this you're doing yeah.

D

A

I mean you cannot use with active span because, because it is your, your span is distributed across multiple functions, so you cannot keep it active, and in that case probably uh you have to you, I mean you have to create the context. Subject: I'll send you the link, I, think that that's something people are doing it uh for the same scenario which you are talking about with. For some reasons, you cannot keep the span active and you have to populate the uh these span contacts and send it across in the HTTP headings.

A

C

Only thing is that you.

A

Have to because you are using your this one, your your you can make a note. You have been creating the text map HTTP text map carrier using your own client headers. That is something you have to figure out how to do it. The line number 37.

D

A

You are using your own client, so you should be able you should be, uh but but.

C

A

Long as you are inheriting from HTTP client, whatever client you're, using as long as you implement HTTP client class you this. This also should work.

A

So you have to come in so just just to summarize, you have to change line number 36 where you have to create your own content.

A

Currently, what.

D

D

Yeah, so currently, what I'm doing is.

D

I'm using the same uh HTTP text map carrier, but while sending the headers what I'm doing is I'm I'm, getting it using the key and then uh I'm trying to send the transparent.

A

Yeah, it should be okay, I think you don't need to do any changes in that. So I think the only only thing is line number 36. You have to create your own contacts and then you have to add another line where you have to set the populate. The span context uh with that. In that context, in in the context which you create explicitly in line 26.

A

and then use that context in inject, in line 39.

D

A

C

I think I can.

A

Send it across this, something which I I know I mean they have been asked. Similar questions asked earlier. Also I can send you the link where we have answered that.

D

Sure sure I think.

A

That will be very helpful.

D

Currently, looking for this yeah, there is one more question so recently uh you had suggested me to uh add a pack and compile using a synchronous exporters, because synchronous was blocking our UI thread.

D

So the question I have with asynchronous exporter, is: uh is it using some some file system or something to write any data or uh how it internally does so the con is, we have a very low memory device and we don't want to increase the memory dynamically.

D

A

It does not use any file for consistency. If that is something you are talking about, it uses the internal memory queue, so definitely the memory will increase in that case. uh Basically, what it will do is that the span uh it's it's the same same issue, which we were talking shortly about the the span. Processor queue getting filled up uh because of the high throughput of the number of spans getting created. In that case, all this we want weight.

A

So basically, the export will not wait for the actual export to complete, but it will return immediately and it will, in the background, in a separate thread, it will be sending the all the currently pending exports.

A

So it's a background thread which would be running to in case of async to do the actual export to ensure that this, the actual span, processor queue, does not get filled up.

D

A

Yeah, but there would be a memory if you're talking about the low memory footprint I mean I. Don't think that that uh you have to you have to do some tests to see how much memory you say you can do for a given throughput of your scenario so because it has to maintain also pending exports in a you know, in memory queue.

D

Certainly I think once we start creating more number of spans and.

D

Sending them maybe we'll monitor the memory and then I can I can get back to it.

A

D

And uh one more small question: last I'll just.

D

So uh to print the debug logs I mean console logs I have added these two statements. One is console debug to true and uh I'm setting this internal uh set log level to debug, so I see that it. It only prints the.

D

Http client data, it doesn't print anything else and, and it prints sometimes, and sometimes it is not printing on the console- is there something I have to do while compiling the so-called or.

A

No, it should not, obviously, by default. If we I mean all the debug blocks should get printed. So if there is any logs which are there in other places, I think that should get printed, but.

A

I haven't seen this issue, but you have feel free to send send to to I mean you can see it. That's.

A

It should not happen if there is something which is missing. I mean do let us know in the slack Channel or in the discussion I. Think probably somebody can see that.

F

You probably file file and issue.

A

Look like the real issue, difficult here, I think in the interest of time context. Sorry, please move to the rest of the things.

D

Yeah I'll stop presenting yes. Thank you thanks a lot.

A

Okay, I think probably the peers.

A

Yeah, this is I, think, probably the one. That's always there we can discuss um this is. Is it open for review now, I I didn't notice, it.

B

uh It is open for review yes and I started to send some comments. What I've reply so far.

A

This is on hold yeah. This I think will take up Mark I. Think probably we are in a good direction for this delegate or AP call. This is something on me, probably I'll. We have to review it. I want to test it and, unfortunately I'm not getting some time to do that, to test it out. Okay,.

B

I took a look at the grpc uh thing also looks okay to be, but uh I've not tested it, but uh looking at the code alone, it sounds good.

A

Yeah, it looks good. Actually it's the right approach to the only thing is that my concern I mean I was um bit surprised because this I use this scenario. I mean in my current test setup and I. Never faced the issue, so I just wanted to test test it once to see what difference it is going, but yeah I think probably this week. I think that since I'll Target to close, it I think anything else. We want to talk in these here this one uh Mark. You want to discuss one.

B

Just to say thanks for the review and if it's okay with everyone I will just melt it down so.

A

Yeah and thanks for the pr looks good to me, I mean much cleaner way of doing it, and thanks for all the functional tests which you have added for this I know it would not would not have been easy. To so add the functional test runs over. This looks good I mean at least from my side. I think it's good I, see I mean other Googles, also so I think at least.

C

A

Have sufficient number of episodes for this yeah.

F

Since this one is good to merge, okay,.

A

Okay, I think that's probably issues we don't. We just had three minutes so uh yeah I added one to allow adding custom, HTTP client, just something I think this was something in line with the discussion which Karthik was doing that to add the custom to allow adding customer HTTP. Client I mean our current design do support that, but somehow they are in Civic. At least we have some checks which probably we have to remove.

A

You know I mean we should be good to do so, just kept it as an enhancement and if somebody wants there is an ask, and somebody is interested from offset Community, they can take it over if this is a valid issue, I think I, probably we don't add, C plus plus 14 black, while compiling in some scenarios yeah. This is his own discussion. uh This I think we cannot take it. uh Std function is not ABI compliant. So probably you cannot just waiting. If you just has any more comments.

A

C

A

Issue um yeah this one I think Mark thanks. Thanks for your suggestions, probably I'll try to fix it by this week and I think it's a valid mix.

A

Yeah I think this is something we have been talking about. You know, yeah, okay, I think we all just short of time. Anything issue which we want to talk today. I mean I, don't see anything apart from these top profile issues which have been created since last week,.

A

To add one thing thing, uh probably that for next couple of months my focus would be I mean I'll, be perfectly working, I mean doing all I mean working in local television, C plus plus I mean as a maintainer and I think I'll be ensuring all the reviews and all any any changes any in general. Any changes required in Matrix I'll be taking care of that, but I think I won't be very active in terms of implementation.

A

Of the things in general, I mean I'll visit spending some time in open Telemetry rest for next couple of months, just to do some implementations and then probably I'll, be back in more full flowing C plus plus.

F

Yeah, probably.

A

F

Is here we can discuss? We do rotation for that to.

A

F

The this meeting- yes, probably.

A

With that I think we.

A

Talk about doing the moderation for this notation, so I think that would be good um in general, so yeah, but just wanted to talk about that. Probably if you see me not doing lots of implementation for next couple of months, it's just that my focus is is right now in uh I mean open, Telemetry rest is somehow lagging behind, I mean they don't have still the stable version of Trace.

C

A

Have one initial version of Matrix as per the new specs but I think I'm not sure how much how much stable that is, and they don't have anything in blocks so plan.

C

To spend some time.

A

For next couple of months, at least in the logs implementation for uh for movement to inventory, rest and then I think meanwhile, anything coming up pressing on The, Matrix in C, plus plus I, think I'll be taking here in general and then it is- and you know in general I think I'll be available uh for any discussions or any all part of the meetings here.

A

So okay I just wanted to call about call of that. Yeah.

C

A

Yeah, okay, yeah thanks everyone thanks for joining, so we won't have any meeting next week so we'll meet next to next week. Have a good couple of weeks. Yep.

B

F

Everyone thank you, bye, foreign.