Open Telemetry Uncategorized, 5 Oct 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2022-10-05 meeting

Description

Open Telemetry Meeting 1's Personal Meeting Room

A

A

A

A

A

A

He took a yeah.

B

Let's see if I can get out.

A

A

You're, just a yeah.

A

A

B

A

A

A

A

A

A

A

A

A

A

Hi everyone, foreign.

B

I think good to see you here for the first time, thanks for joining.

A

B

uh Let me share my screen uh and probably after this position,.

B

B

Here, and probably uh just for the interest of a time forward, I know it's a bit late for him. uh Should we just go for the locks SDK and have a quick discussion on that yeah?

B

That should be fine, and let me open the discussion in the.

A

B

I think, uh just to start with oven. Most of these uh I mean the way the the points which you have summarized I think they look good. uh The only concern I had was with the recordable interface whether we should remove that or not.

B

We have been using that in as of now, we do use recordable interface both for traces and uh for locks and if we can continue using it.

B

I mean the even the trace, API talks about uh readable span and read, write spans. We don't really implement. We just haven't Span in this. We just haven't spanned object and then most of the things are happening to recordable.

B

So if we follow the same design as we are doing for span and just use the recordable interface recordable would be basically how the exporter looks want to how the exporter want to serialize the data. The logs data. Okay, do see your comment here.

B

Specification says that we have a log call in the API API and create uh read, write log record which derives from.

B

B

Okay, let's open the so we just specification once again.

A

B

B

B

B

The logger will have the emit method, which is good, which is going to emit a log record.

B

Will be called as long agent.

B

I mean I was just thinking that, instead of having a log report as a object, if we just have all the fields which are required to be emitted as part of the argument to the emitted method, something like which we are already doing it.

B

There's something like a logger emit event and that that or logger emit that will basically have all the fields which are required to be emitted. Instead of creating a new log record object which will store the data I will just emitted using this.

B

Or my only concern is that we should not create any intermediate storage for all the uh I mean all the data which we have to record I will know which we have to basically send up send out to the exporters, so they should not be an intermediate Stone Age.

B

If using this design, we are not doing any intermediate shows I think we should be. We should be fine.

B

As long as we don't have any intermediate storage of the all the data which we have to log, so if this design looks fine as long as there is no storage, so this just.

B

um Record this.

B

I mean um as long as this is this is something this. This is some. This is a format which is used by exported to serialize I think we should be good to to have it uh in property, otherwise, I think we should not.

B

um We have to change the design or keep it the way we are doing it. As of now.

B

But yeah, let's, let's continue discussing in this, uh let me try to understand I think how how how this is going to work and probably like. Let me write, let me write the design which I have been thinking and we can talk. We can talk over this discussion in this.

B

uh Will that be final event or if you want to share something.

B

So so so let let me let me write the design where I was thinking and then probably I think we can discuss uh further here and then, let's, let's, let's talk about this further in this discussion.

B

Thank you thanks for thanks for proposing and thanks for summarizing all the changes. I think this looks. Definitely this looks in the right direction. Probably once we once we finalize the way uh we are going to uh send, send the logs from from processor to exported I think we should be. We are done with that. I think we can find language here and then we can create. uh We can create the tasks and split it across So yeah. Thank you.

B

Thank you for this um yeah and okay, and probably we can quickly go to uh if there is no other comment on this I think we can quickly go to the rest of the discussion. You know uh in the specs changes we have open, Telemetry specification release, 1.14 1.14.

B

I, don't see any probably, we can just quickly see if there's anything which we have to do in.

B

So it has a change in the default buckets which we have already done, uh and this is open. Matrix I think we don't need to do anything in this environment variables for batch log record processor. These are optional. So probably we can this exacts new semantic attributes. Probably once once we take once we rerun our semantic attribute generation, we should be able to get them.

A

So we need to align on semantic conventions and.

B

Yes, we need to know.

B

It does has Hotel SDK, disabled environment variables, uh I mean I, think we discussed it last week. This is I mean all the environment. Variables are kind of optional, so probably if you can support it without much of the changes in API that is SDK we can, we should be doing it, but I think.

B

I think I'll leave it to you. You Mark I mean how do you want to do for this.

A

B

Yeah there are a couple of issues have been raised for more of a clarification needed on export and greeted configuration. I think there is a very thin line of different as of now what exportec can configure and what reader can configure. This is for the Matrix, and probably it needs more clarity. uh So there is aggregation. I mean how default aggregation, whether it should come through reader or through exporter.

B

There is a the aggregation temporality who is going to who is going to configure that as of now in our code, we are doing it through exporter, uh but I think the specification is bit not clear, some somewhere. It is talking about exporter, to configure it from where it is talking about reader to configure it. So it needs more clarity and I. Think I'm, just just probably keeping an eye on that. If we need to do any change on this, I mean, ideally, it should be with exporter.

B

That's what I felt, because exporter you know I mean is- is more aware of what kind of temporality the uh the downstream systems are going to support So, based on that it can configure that accordingly reader may be hand taking care of multiple exporters. It may not know what all input is. They may have different temporality or it may not be aware of what temporality the X the downstream systems are uh supporting. So probably exporter is the right place to really configure that, but I'm not sure I mean leave it to the specific him.

B

As of now I have I'm not commenting on that I'm just waiting. What exactly is they're going to do that? So both both these are basically dealing with the same issue.

B

uh Magic exporter complication extend beyond the scope of its ability. So again it talks about the same temporality. This one says that this is a job of reader, not the exporter, but yeah, probably I think just keep an eye of that.

B

If it is goes to read it, we may have to do our modify our exporter and read it in the Matrix to take care of this, and uh this is another issue which is uh this is a PR I think this is probably having more discussions right now in this clarify the attribute support based on the stable wire protocol definition of attributes.

B

Okay, this is more discussions are happening on this and this. So this is basically whether we support the tested attributes or not. Both I mean at all the level at this, this parent traces, the traces and the knocks in The Matrix, whether we support it or not. So in our current implementation, we don't support nested attributes.

B

So if we just go to I mean electrically, we can go to you know: DPA dude,.

A

B

Yeah, so we don't support nested attributes, so the value types can be deprivative types or it can be I think this would be mostly array or vector, or something and yeah we don't have an estate attribute support.

B

So if it is going to add that support, we have to see whether we need to do any changes or whether we should do or not and or we have to flatten their flatten all the attributes and send it across how how do we want to handle if we are getting that, we really need to support anything in this, but yeah. That's that's. Probably we need to take here uh if, if any changes happen, Logistics we already discussed uh just to clarify again I. Think Oman p is uh looks good.

B

As of now, probably we can we'll discuss it further in the discussion forum. Probably probably I, think the right approach, I'll I'll put the way. I was thinking, and then we can again discuss on that thanks once again for putting putting summarizing all the details in that.

B

Yeah we do have to discuss on the meeting time. Probably I'll take it afterwards. There have been lots of proposals of the time whether we should do it, whether we should not do it. I think we'll have a quick discussion afterwards in the end for the PRS. Let me know if any anything else we want to talk here. Otherwise you can quickly go to the previous.

B

B

And I I'll go to the agenda. I'm interesting discussions are there so, let's, let's quickly discuss pics, more sea lion and GCC warnings, okay,.

B

Mark you like to talk about this sure.

A

So I'm still in the process of cleaning up every morning, and so this is a new barge. There is something like 70 file change for this one, with a lot of things detected by silang and GCC and I'm doing that. Incrementally one batch at a time until the Vlog can be totally clean so expect that there will be even more like that.

A

B

Words are very.

A

Very simple, so.

B

A

um I went would like to have this subtitled on. Is it already on or.

B

Subtitled, what do.

A

You mean for the meeting.

B

Sorry I didn't get you.

B

Subtitle I mean I, didn't get you sorry.

A

Yeah, you can go.

A

A

In the chat he's trying to turn.

B

On it, okay, sorry.

A

B

I am, oh, is it does it support subtitle, I'm, not sure.

A

Gonna be awesome.

A

Okay, so it's not live.

B

It's yeah, it's not live, which is kind of sad. We should switch to teams. At least teams has a life translation.

B

B

To me as of now, but let's let's go through this and there would be more more changes coming mark right, I'm, just just to understand how bad we are in in terms of the code quality. When we are talking about the warnings.

A

uh There is quite a lot. The thing is, there is so much when he's creating so much noise that it's hard to fix everything in one one shot, so you have to fix some to to get more ones.

A

I had said this, and maybe between merge the pr with the judge the warning enabled, but only enabled for some branch, some feature branch, and then we can raise PR against that branch.

A

Just well, it's uh I'm I'm, basically using the the pr band for that, because it really it does the CI jobs in GitHub. So I can see for the new CI. The warnings I mean it's. It's a tedious thing to do with just wanted to help sure you're, not only you're, not the only one lose.

B

Okay, you're saying kind of having a feature Branch with warnings and posts and then probably do all the pr as to in that branch and then finally, we'll merge that Branch to the main branch.

A

Well, it's it's going to be a big merge and because it's.

B

A

Code so many places there will be a lot of noise issues, so I think it's okay to fix and move to to the main branch.

B

A

B

And Mark, just one question like: do you think we should have some line like what extent we are going to fix the warnings will are all the warnings really really required? I mean these are looks, definitely a valid case to be fixed, but just want to understand like uh is it that we have to fix all the warnings which are coming, which are supported by DCC by enable everything or just.

A

B

A

I'm only focusing on uh what is in One X transfer file; okay, fine, fine yeah! Thank you. This should be.

B

B

Okay, yeah and.

B

B

Okay, this this looks: okay, I think we should be able to merge it, and there was owent has a question on this.

B

The name of the I mean the format is not fully aligned with the RFC, which is a valid concern, but I think if everybody is doing it in this way, all the sticks are probably worth doing this way and and The Collector is going to read it and it is able to recognize it. I think we should be good.

B

I, don't see any comment coming from The Collector team, where I did put him here, but he has not replied so we'll wait for some time and I think it's okay to merge it. We don't. We won't fix it from our side.

B

um We should we want to fix something which there, which.

B

It's okay to keep it the way collected team is asking: let's not try to be be RFC compliant for now. If something is required, we'll fix it afterwards.

B

A

It will be triggered to.

B

Fix at one point so yeah yeah I will do that yeah. In that case, I mean I. Don't know it's good to be in compliance with whatever species.

B

Picked up the rule gauge matrix, it's a small change. There was a I mean kind of some some problem. We are not maintaining that timestamp was not getting updated ah for the observable gauge or for the last value aggregation. We were not updating the timestamp at the start. So uh during aggregation during merge, it was not.

B

It was not able to figure out, which is the latest data points, so it was not doing the proper aggregation because there was no timestamp, so small change, probably I, think if I mean it's on, if you can just have a quick look, I think you should be good, nothing much nothing much in this most of the changes coming from the test unit test which I have written, we didn't have it in a test score last values, so I just added it added that unit is.

A

B

Add a maintenance mode in cmake.

B

Okay, I haven't been following it for past couple of days, but.

A

So basically, the this is to get a build in CRA. uh Well, there are two things: first in CI, we need to build uh what tracks we're wanting and as a developer, it's very annoying to fix code, push the 2D tab to see a warning in GitHub and and do a commit to change it and go again. So a developer also needs to locally do a build uh with the warnings.

A

So this new mode in cmake allows that if you build in maintenance mode, it will basically do a build with running all and running extra and so that any developer can test locally. uh If a code is clean or not- and this is reused by the CI on GitHub to do the same- build when we push on the when we push some some PR.

B

A

So I I've added the three platforms on CI one for C dang, one for VCC and.

B

A

For visual studio and Owens, provided me the flags to use for visual studio.

B

Okay and and these these CI tests will keep on failing till, we have on the morning fixed right.

A

A

So we have two choice: either we merge this thing now, but we will see this failing uh for everyone or we wait until the code is cleaned and then we can merge that and then we will have clean build on CI um I, so this process is likely to to iterate a few times, because there is warning in this grpc that needs to be silenced by our warnings in frift. That needs to be silenced. Things like that, so uh I just put that PR in draft mode, because I expect a.

B

A

Of tuning in here and there in make files to adjust.

B

Okay, right, that's, okay, I think, probably once once uh we are sure that all the warnings are fixed and silenced. I think we can probably was this year.

A

B

There were tank tanks Mark for your vote on fixing all the one yeah just pending for a long time. Yes,.

B

And arithmetics example of Internet yeah I think this is good, probably to merge. I, don't have any more comments. I just want to understand. As of now, this is disabled. By default, right I did had.

A

B

A

Some background that supports this.

B

A

Just to create uh to have some some example with some back end.

B

B

Yeah I mean I was the reason why I was asking. Is that? Because this is as of nostril experimental, and it clearly says that it should not be affecting the performance of the novel matric generation? So if it is disabled by default, follow I think it should be good. That means that will get the proper reasons in.

A

B

A

B

Yeah and we don't need to put it under feature flag right, it's it's just because it's disabled by default. We are good, you don't need to put it in the video yeah. We.

A

Already had that that one, it's been called No up something yeah I.

B

Saw that and so I just wanted to confirm.

B

Okay I think that's, okay, probably I think we should be good to merge it uh I'll do it only after after the video feel free to do it, I mean you can do that.

A

B

And let's go for the issues, uh probably because you can see: okay, let's put it yeah. This is already we have discussed uh bill, bondings yeah, reported by sea land. This I think we already had a discussion. Rd oxygen, Build, Your Mark, thanks for using this and assigning it to yourself handle gaps and reset for racing instruments. Here this we have to probably support it.

B

As of now, we're not supporting I mean like a sync counters, are monotonically increasing values, so it's possible that some that the instrumented code resets that and starts generating new measurements started starting from initial value, which means that it will break the observable the the async observable counter. So we should be able to detect that and we should start with any research happening. We should region start regenerating The Matrix from that point onwards.

B

So as of now, this will give incorrect results if we don't detect any certain reasons, so guys just put it in the Milestone we need to see without towards here.

B

Space 2 express it 64-bit integers. Yes, this is something we need to support it. uh It's just not 64-bit. We should support both 32-bit and 64-bit. Rightly. We should be supporting 32-bit 64-bit long in all that Architects at the API level. At the SDK level, we are going to we support 60. This informatics are stored at 64-bit, so we should be good enough for that with that, but at the EPA level, I think we should support both 32 and 64.

B

to start with, I think we should it's good to add the support for 64-bit, and uh so you can change long to you to uh in 64 and then probably afterwards, you can add it for 32. Also.

A

B

Yeah I just kept it unassigned, probably I can take it to if it.

A

B

B

I think it should be this as of now I think it's good, it's good to change along to in 64, and then probably we can see letter to add 32 on this.

B

Thank you for taking it.

A

A

B

A

The priority will be P1 yeah priority.

B

Should be B1 here.

A

So documentation.

A

B

Okay, I think this is P2 but I think about something different. You need to fix. It.

B

B

Issues created as of now by the documentation team, but probably I, think we can start working on that. One see if it is there.

A

True, it's uh it's in the list which is ever going so.

B

Yeah 100 partial success response. We need to support this also sometime and then start documentation yeah. This is more of a documentation and property. Just I don't see this this, sir, some more more with the Matrix and I think we can talk about the Matrix milestone.

B

uh I think we should probably 15 does not look feasible to me. Just correct me, I am would like to change it to say 29th or something or probably.

A

B

I think it's, let's not hurry on this I mean it's probably fix all the existing issues and then.

B

This is something which is I, think we can close it and I'll leave it to Sr I mean if there is anything open on this. Probably you.

A

Can no I'm just going to merge.

B

It yeah yeah and any.

A

B

You feel it's something is missing as of now probably you can open an issue and and take care of that anything anything further if it is still missing. One exam, filter and Reservoir.

A

It's it's just increases the example and we don't have I.

B

A

Anything supporting that so.

B

Oh yeah good point yeah even I'm, not aware if you have anything to sub in the back end, supporting it your provider Builders, to avoid exposing that this is would be done early once we are near to the completion of the API anything. The number changes happening.

B

I had support for pull Matrix exporter now this is something yeah. Probably you can have a look into that I have to see if any of the existing E6 are supporting. It. Probably that's also a good thing: if it is there and I think we can, we should also do it I'm, not aware of that uh add support of exponential. This is I, think we can move it out. I, don't think it's very important as of now yeah. This is something I have to take care clean up a whole Matrix.

B

At the end, this I I'll just do a more resource on reportable Enterprise between storage and exporter, documentation, yeah, you have to do it and this to explain it 64-bit and handle caps, so yeah I think probably it's good to move it for in a couple of weeks and then take take a call uh in couple of weeks, but, let's, let's I, think I'll be spending more time now in Matrix. The work which I was doing for contract is mostly done, so it should be good.

B

I mean the work which I was doing more internal work, which is mostly done so I. Think I'll have more time now to focus on this and uh yeah. That's that's all I think there was uh yeah. Let's go quickly to contrib I. Think owent has raised a PR which looks good I have gone through this.

B

uh It looks good to me, but.

B

Yeah, thank you. I think this is, you can should be good to merge. It.

B

Thanks so much for this PR uh and yeah I, we have it I discussed more for internal, but I. Think Tommy is not there today. If I will work with Tom I am working in apart from that to create an instrumentation library for Matrix more for process Matrix. As of now, which is uh going to collect all the process related Matrix in terms of CPU disk memory, utilizations and and this this is more for testing properties, respond to C and to end.

B

We can use it for any backend systems and try to see um you know. What's what's the performance end-to-end performance which we are getting with our Matrix is speaking, so this is still ongoing, but yeah and okay, let's quickly discuss about a time, I I'll be more I'll, just uh probably three to four minutes uh so meeting time, uh I mean it's just that I saw all the comments.

B

I mean, probably uh it's just that I don't really see good or in general, it's not really good to have the meeting times which, where people have to join night at 11 11 o'clock at night or 12 12 a.m at night. That also goes through with our uh I mean Monday meeting, which is at one PM, uh PST I know you guys Assan and Mark.

B

You have to join late night for that, so probably I think it's good to have meeting times in general, where people are people are able to join in their normal working hours. uh So the.

A

B

Can also be happy.

A

B

Difficult to have yeah I know I know for you. It would be difficult for you to join I mean with normal working hours, I'm saying 6, PM, 7, PM or 8 P.M I mean anything going. Beyond, 9, pm or 10 pm is stretching bit too far for anybody to join. It's not setting a good example.

B

If we are doing that so that yeah, that wasn't concern, and even if we move it to 8 A.M in the morning, 9 A.M meeting to 8 am and I don't think it will help owent, because once the daylight saving time ends, it will again become 12 a.m at night for a wind. So even if we try to keep on it to 8, A.M I, don't see that will help Owen to join the meeting.

B

A

B

A

Think Rian riang uh I, don't know if I say it right, but if proposed to have uh basically uh dedicated meetings to uh to look at a specific point. Instead of using the regular meeting time.

B

If we have to talk about a log, SDK and API I still feel for locks, I think we can follow. We can have the discussion forum I, don't think there are lots of conflicts which we have to resolve.

B

So probably we just need to move, have a common understanding of how we are going to implement the sdq and I think we can discuss it on the discussion forum, but in general, if we have to do uh any such discussion, I think we can have and ad hoc meeting with with these specific participation, we can have a discussion on that, instead of changing the meeting time. For that I mean, if you know, if you remember ehsan, we did change the meeting time from 10 a.m, to 9 A.M for uh for the Cisco team.

B

They join initially for a few a few of the meetings and then they stopped joining, and then we are stick to with that time. So lots of changes I think not not setting a good presidents. If we are going to change timings every now and then uh having said that, I think still the 1pm time does not look good to me specifically for you, son, you and Mark I mean I need to feel bad for having that time, I'll be open.

B

If we can move that time to 9 9, A.M or something that would totally fine for me.

A

Well, for me, it works because on Monday I'm I'm actually available and it works, it works well, but uh sure we can. You can move it earlier. If we can.

B

Yeah, probably we can discuss about that time. I! Think in that discussion. Let's, let's discuss it, discuss it offline and I think it means we all agree to move that I think we should be moving.

A

B

I don't see 1pm really suiting in either Asia Pacific or european region. I, don't see it suiting any of the time zones. Why do we have one PM I? Don't understand it's been there for long, but yeah, let's, let's yeah, let's, let's make a call on that. We really want to do it. Foreign.

B

Feel free to discuss, if you want to do there are something which we have to talk.

A

B

Okay, so yeah it's nothing else. Probably we can. We can close a meeting. Sorry about this I think I I have a really hard stopper I need to okay, who got for some meeting. So thanks for joining us. Okay, let's, let's meet yeah thanks.

A

Everyone next week, next.

B

Week we will meet as as as per the current time and then we'll discuss if we want to change it to you. Thank you.

A

Yeah thanks bye.