Open Telemetry Uncategorized, 30 Mar 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2023-03-30 meeting

Description

Open cncf-opentelemetry-meeting-3@cncf.io's Personal Meeting Room

A

B

Whom did you do that.

C

B

We can't hear you not sorry.

B

B

I can't hear you anyway.

B

Hello, hello: there you are okay,.

D

E

Good morning,.

B

Good morning, Stephen.

D

C

We are probably all.

D

D

D

Administer foreign.

D

D

Okay, I have this well.

D

So I think we are all uh we're not he's not joining today. So uh we are a usual suspects suspects here. Yes, uh yeah, uh okay uh for last meeting actions, uh uh yeah um yeah, we talked about the spec version. That's now in uh inappear it was already it's already in the library. uh Uh Kevin was gonna, add diversion header to the HTTP or TLP code.

B

All right so I'm, sorry.

B

um I've I've not done that yet I've been working on the other issues, so I'm I will get to that shortly.

D

B

So, but that's the only exporter we need to put it on is just to set the user agent when we're talking HTTP to uh an otlp collector. Is that correct.

D

Yes, it must be added to the request.

D

F

Think so it'll be to each um signal uh alternative.

B

Exporter, of course, okay, okay,.

D

uh Yeah I had to update the spec version because we, the the docker image we were using I, don't you were not here Bryce, but the docker version that generated the it's called a command same code that generated everything, uh don't work on M1 Max, uh so that had to be done with another computer. uh My computer, now it's an M1, so I couldn't run that and I said. I will change that with another computer, but I didn't have time for that. So I will I will add that also so.

C

D

And I think that um vinod was gonna check. uh What was needed for that, so that I think I have can work on a Docker on top of im1 so yeah, so we can run it uh nor use more, usually so yeah okay uh Matrix will write uh uh yeah. This was.

D

Yeah the we added your comments in the last meeting uh about the metrics of right uh uh uh price. uh Any update.

F

um uh Let's see yes, so uh I think I did miss one thing, which was the uh the actual um exporter for the stable metrics, which I added last week or maybe earlier this week. I can't keep track um yeah.

A

D

Yeah, it's all it's a.

F

In a draft PR, but it needs a lot of testing and you know tests and that sort of thing, um however, I'm going to take a break from it, because I'm a little burnt out on it. If, if nobody else picks it up to do the testing um I'll, probably pick it up again after a month or so we'll see.

D

uh Yeah personally, I got approval from my manager to spend more time on open Telemetry awesome.

C

D

Know that I was limited, but now my the manager who said that is.

D

Good so now the chain uh has changed a bit, and my mortgage manager said that I could uh take some time for it, so yeah I can I can work on the test for that price, cool.

A

D

Okay uh uh yeah, so it's in the metrics brand too.

F

Yeah I think there should be a PR open in the open, Telemetry Swift repository with with it all in there.

D

Okay, yeah, maybe.

A

D

Can also maybe yeah check that and yeah and work on that.

A

D

I will try to to add that.

D

Yeah, probably.

C

D

The existing tests will make sense, or do you think it won't did.

F

I say that again.

D

uh Do you think that adapting the current test that we had for the previous spec would make sense.

F

um I think that things are significantly different. There might be some overlap, but uh I think overall, it's it's going to be um just yeah totally new stuff, yeah.

D

Yeah just to know okay, uh more uh uh Linker, error issue. This was about separating the otlp libraries so separating uh the grpc with the agency versions, because the erpc has lots of dependencies that are that generate problems for many people. So and the idea was, if possible, having a common code Library, that's internal, somehow and used from outside something like that. Right.

B

So I I spent some time on this this week and um it's it's fairly clear I either need to know more about how we build projects in Swift, or maybe someone else you picked. This up I mean I, basically was trying to model. What's in the pack, the two packages that Swift files and I sort of got nowhere with it. So I'm clear, you know and I read some of the doc too and didn't find anything that I was clearly um missing.

B

So uh yeah I either need someone else to pick it up, or at least um maybe give me some tips on how to get started because I got I sort of got to the you know, throwing stuff at the wall stake and stage and couldn't make any more progress on.

D

It yeah I think uh yeah uh uh anyone, uh maybe Carlos or Stephen. uh Would you like to work on something like that.

E

uh uh Stephen here yeah, certainly I I would uh I would enjoy that um I'm not going to have any times in the next month or so. Okay,.

D

E

uh Yeah, it's gotten a little crazy.

C

F

Carlos, your audio is uh unintelligible. Unfortunately, yeah.

D

God's father, like uh audio really.

D

G

Any any better now there.

F

E

G

E

G

Okay, I was saying um probably I'm in the same boat as Stephen for the next month or so um and interested, but not it would be a little bit of a learning curve for me as well.

D

Okay, okay, uh yeah, so okay, yeah I, might try to to to do something without them.

D

Let's see if we can uh yeah move that around yeah or maybe ask for some external help, if possible,.

D

Yeah I don't know uh yeah, because I have the next topic. The first topic here that I have been working on. We probably take a lot of my time for.

B

D

B

I can I can try to read some more of the doc and see if I can figure it out, but you know so far. I've sort of just run into uh yeah.

D

B

D

B

D

Okay, uh so for the rest topics uh uh from last, no more things from last meeting uh the first topic I had uh brought here was uh URL session. Instrumentation is not working for iOS 16. uh Bryce, uh open an issue on the slack channel. uh I think it was on Tuesday uh that something was not working for him and yeah. This tldr iOS 16 changed the way uh URL session work when using async away methods from URL session.

D

So all the uh methods that we were monkey patching or Switzerland for doing all the logic that includes not only logging, the request, but also ingesting our Trace headers and that stuff are not working for iOS 63 we're using async away. So if they work for iOS 15 with exactly the same code, but not for iOS 16. uh So that's a bit.

D

uh That's a big problem. uh I would say uh so. I have been working on that since uh yesterday, after Bryce told about that it was originally my code.

D

um My I have I I had two ways of uh approach it. One of them was some new hidden methods that URL session was showing when running on iOS 16 that had similar names but had an underscore of before them. uh I have tried with them today and Away uh they yeah.

D

There is no way to for that to work. As expected, they somehow appear to be called by some methods, but not with the proper parameters.

D

So it was a big failure, uh at least for me. uh The other approach I started testing. Yesterday, a bit and I uh my plan to continue is uh there are two methods that are still being called. One is the resume method for your for the URL session task, uh uh even when there is no attack anymore, that resume is being called.

D

uh So maybe we can move some of our code there uh I, don't know if that something I must check if I can access the URL session through the URL session. I can access the request and modify something there, because I think that that's a modifiable request the one there.

F

uh You might have to recreate it and recall it or something yeah.

D

uh I don't know, and the other method that we have, uh that is being called, is the only method being called because uh all the other delegate methods are not called uh I tried with a delegate. Also I wrote all the delegate methods and just uh the one only call was the one, the metrics uh one that is like uh reports, the network, metrics uh uh net I, don't know what I can search for that method, but I mean the one that returns a network metrics for the request.

D

That method is still being called, so we can get information about the status of the response about the duration and that stuff from that method.

D

So I think we can get some information there create some at least not expands, not life expanse, but after the request spans, at least with things being called, but uh not sure if we can ingest new headers to all the requests, so yeah I still a lot of work to do uh with with that. So I I continue working on that. As for me, it's a it's a P0 on my site so uh trying to have something I hope to have something uh at least help working for next.

D

Next meeting, oh next meeting, yeah next meeting that I want to be here because it's on holidays in Spain, so yeah I have a bit more time, uh but yeah I am working on this uh as my first priority, uh because for me it's uh is yeah. It's a it's a big issue not having uh instrumentation for networks currently and I, can expect. That's also an issue for anyone. You see open Telemetry, also but yeah uh so yeah. That's my Approach I, don't know Bryce.

D

If you have another idea or how to act, uh I have been taking other libraries. Sorry I have been taking other libraries. uh That also did this Switzerland method uh there and I have seen that they have did habitious with the with the problem uh that that's not gonna. Work I also read about one that said that they had not this issue. I had tried it and they have the same issue so yeah I, don't know if Switzerland is the way to go.

D

F

Yeah I, don't know what other option there is really.

D

These are something like uh URL logger from Apple, but that doesn't allow to modify the request. I think right. uh At least it was not possible in the password and I brought this code for uh method 37, so yeah I, don't know, that's a big hole. We have you.

F

Know yeah I mean yeah I've run into I. Think I mentioned this to you in the in the chat, but I've run into this sort of issue before where, like the data tasks that are getting built by the nsurl session, are like some sort of private, uh instant or, like you know, private object with.

F

You know that don't call the original methods, but they implement the the you know, interface and you know, by creating those objects and then accessing them through the um runtime you can get, you can get access to those methods and then Swizzle them that way. But it's not an ideal thing. I, don't know. If that's even the same sort of situation that's happening here. It sounds like they're calling separate methods altogether, um as opposed to the ones that you expect them to be called so.

D

Yeah, my main concern is that they are using some native Swift Code.

F

D

uh uh Because it's only happening with the async thing. uh I also checked the uh URL session that it's open source from Apple on the Swift core libraries, but it doesn't have any async await code there. So.

F

D

It's an old URL session code, so yeah I. It's we're a bit a bit blinded here, yeah.

G

What about an S URL protocol as a way to intercept requests as they're going yeah.

D

But but I think it doesn't allow you to modify the request, the outgoing request for ingest for adding headers, for example. That's it okay.

D

Modify our and that way, maybe they change that now. I, don't know.

G

Yeah I, don't know it's a dark corner. I haven't looked into.

D

D

Okay, yeah I will I, am gonna work on this full time, so I hope to have a solution or yes or we are dead. Basically good.

F

D

I will include if we cannot do it automatically, we will need uh cast users of the library to add things themselves. Yes,.

G

D

Not good disabled, because yeah we need magically added was was really nice, uh especially for for uh for other libraries that they're custom reduces. So if you have a uh intermediate framework that uses open Telemetry under the hood, uh having that for free was was great. If you have to make the user the final user of your library to use things that that can be a bit difficult, I, don't know uh yeah, okay, uh next topic, Bryce.

F

Oh yeah, so um this was just something that I noticed when and it could be just the way that I'm doing my instrumentation for view controller stuff, but um it seems like active spans once they're completed, aren't removed from their from the uh the active span pool. Is that part of? Is that something that should be happening? Or yes,.

D

Yeah and uh should be cleared from the from the from.

F

The context from.

D

The context it should be done automatically: okay,.

F

This I think this is my fault, because um I'm I'm adding a span to a lot of different contexts at the same time, depending on you know like that. This is the problem where you have view controllers loading and for some reason, like one view, controller is in one context and the other view controller is in a different context. Even though they're, like related, you know, they're still being presented on the screen together. So I add you know my span.

F

I want one span to cover all of them, and so I set the active span context for one view controller and then, when the other one pops up and there's already an active span for the view, controller, instrumentation I, add it to the context there. So anything that gets called from that context also gets added to the active spin, but when it completes it only it only removes it from the active span that or from the context that it completes in rather than all the contexts that it is actually in so um I.

F

Don't know if this is something that needs to be extended in the in the in the actual uh tracing stuff um or because I can clean it up myself. It's just um it's. It's just uh I wasn't sure if, if I was to you know, if it was a mistake in the in the um span code or not, but.

C

It sounds like.

F

It sounds like it's my fault.

D

D

Yeah when it gets removed from the active span uh yeah, but it's a context related yeah.

G

D

It only clears for the current ex yeah.

C

D

If it's also the active Span in other contexts that will probably not be removed, yeah uh yeah I think if you can write a test that fails uh I think we could do something, because probably we can I mean I, think the current code. I I, should check it, but the current code tried to get from the current context right, but maybe we can do the opposite and go and search for all the contacts that has that span yeah and remove them.

F

Or like maybe provide the span with a list of contexts that it's active in in that way makes a little bit easier to clean up at the end of the day, um but yeah I'll see if I can I can add a test that exercises this issue.

D

Yeah, it will be, it will be great.

F

Because it yeah because it see it seems like yeah- maybe maybe it's against spec- to have I guess the context idea is a iOS specific thing.

C

F

Yeah I mean maybe this is something that we just have to Define like. If a span can be added to multiple contexts as the active span, then it should probably go through all those contexts and make sure that it's cleaned out after it completes.

D

D

Yeah, it removes for the current yeah yeah, because it only takes this uh con. It doesn't okay, yeah.

D

Yeah we could yeah I think we I can also take a look for that. Other um I I had not.

F

Yeah I mean it's: it's not super critical I could write up an issue just describing the problem as well, um and just so that we can track it because I've I've come up with a temporary solution. Myself like I, just make sure that, as the context like like I have access to all of the view controllers, as they're finishing I can just say like make sure it removes this active span from this context as well as it's completed and.

D

That's and that.

F

Solves a problem for me, but.

D

F

Does seem like we should probably, if it's possible to set a span in multiple contexts, we probably need to make sure that they're cleaned up as well.

D

Yeah I I, yeah I.

D

Yeah we yeah remove objects for week. We can probably uh try to find this object. Yeah.

F

A

F

Might be an expensive.

A

F

Might be pretty expensive because you know it's like you know, in squared kind of operation, to search through each list.

D

Yeah, that's true, uh it could be a bit uh I mean it.

F

Wouldn't really have any contacts.

D

That will probably be a nightmare, but.

A

D

I mean yeah, probably is the opposite church that we need to do, but uh I think it can be I I. Can it's just modifying this to take an account yeah.

C

D

Activities because we are just getting this one, uh so uh probably we can iterate here for all the for all the yeah. We can yeah that that can be an option. Yeah I had never thought about that option, but probably instead of searching for the current activity uh we can search. We.

F

Just look through all of them.

D

The the value in these cases are spam, but it's also usable for baggage, yeah.

F

D

So yeah I think we can uh yeah, but it would be yeah having a text will improve it, but.

A

D

I think we can fake it also.

F

All right I'll create an issue to start with, and then we.

D

Can yeah? That's that's great yeah uh Yeah. The code is very simple in reality. Yeah the tricky thing was uh accessing this.

F

D

D

Okay, true uh any other topic. Anyone.

E

D

E

Nothing for me today, sorry.

D

Okay, no problem.

D

Okay, then I think we can. We can here all.

F

Right have a good weekend. Everybody.

D

Have a good weekend thanks.

D

Thursday uh uh bank holidays here.

E

In well, that's right! That's right! I did hear that. Yes, sir okay enjoy yourself have a good time.

F

Good day off, yeah.

C