.Net Foundation Design Reviews, 7 Jul 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: GitHub Quick Reviews

Description

00:00:00 - Approved: Improve Activity API usability and OpenTelemetry integration (Part 2) https://github.com/dotnet/runtime/issues/38419#issuecomment-655054487
01:44:16 - Approved: Configuring request options in Browser WebAssembly https://github.com/dotnet/runtime/issues/34168#issuecomment-655067469

A

All right so today we are looking at three issues and we will start with the open, telemetry integration, every probably look at the design document, instead right rather than the issue here, because the movie doesn't do much for us. Oh.

A

And now we have this beautiful thing right.

A

I'm only two clicks away.

A

Perfect all right.

B

System Diagnostics activity just to to address the scenarios of inflammatory and to be able to give the libraries and applications the ability to publish tracing just depending on the PCL. Only.

A

B

You don't have to.

B

C

B

Device is used in implementing open telemetry is Decatur, I mean just to prove that, like like say, VI is missing all open trauma tree scenarios that we are addressing here. So this these changes that we are proposing today is just the feedback we got from from this activity about how, when, when the API is used in the telemetry is decay, and you like what kind of scenarios that it's kind of like locked or it has to it, need to be like enhance it. Just what we see now is that it ends up in cavity.

B

So this is an introduction. I mean like you know, we can go to the list now. I mean we can address every every point, so the first one is activity. Texts, third context is one of the new types that we exposed in Lausanne in previous previews and activity context. If you remember it contains, like dress, IDs span, ID flags and state- and you know telemetry, they have something. It's called is remote, which is telling if, if this context is created from a remote parent or it's or it's not so is remote is not something new.

B

I mean we were aware about it before, but there was some like debates or questions regarding it, but after discussing it with altimetry guys and like the inside about like this is very important to have, and we need to have it in our egg, your eyes so basically I'm, just adding a new property is called, is remote, which is it's just boolean, and also I adding I mean like we have the original constructor I'm, adding option and parameter at the end is called, is removed equal force, so this is basically just to support.

B

Is remote property?

B

Any feedback on that.

D

This constructor already released there is a new. In fact, it's released in the previous.

B

Preview but yeah, so we adding I'm expecting like literally, will see also we have some kind of breaking changes, so we're changing a couple of things that we already exposed so we'll come to that, but so I'm expecting like you know. This is kind of breaking change from the previous previews, but there is no breaking change from what we released before so.

E

There's there's never been a stable package with these types in it.

B

D

Do we anticipate adding more parameters in the future.

D

B

Is not I'm not expecting like we will and what.

F

Do even have those are open. Telemetry consumers use this provider for.

B

This this actually it's it's kind of like you know, while Hawaii implementing something like sampler or something like this. So we send the parent parent context 2 to 0 I mean to the listener and listener is used by the sampler and assembler. Maybe can decide to take some decisions according to that if the parent is really and what context or just a local context,.

B

Any more questions regarding that I.

A

Mean my only concern is remote as it sounds like the innocence itself is remote, which is not the case, but I also don't have an idea for how to name it any better.

B

They are using exactly the same. Are.

C

We implementing open, telemetry API is directly or a return to make a dotnet version of it.

B

We are not strictly implementing directly, but we are trying to be as much as much as we can with. So we are trying to reduce the confusion between, like you know, somebody using automation or somebody using easier I mean like they can easily understand or live between emails.

A

Okay I mean we generally our stands gently as we don't just implement specs why we generally try to make bcl api's that are self consistent with the rest of the platform first, because the problem is one day we don't like the specs. The problem is that specs are fundamentally not consistent with each other either and implement more than one external protocol or standard or spec, right and I. Think the question now becomes, if we do this over time like what happens to the BCR alright.

A

So that's why I personally don't care that they would call it as remote. The question is just if I see is remote as that developer. Does that make sense to me, but, as I said, I, don't have a better name for this either or like it's just a remote reminds me of remoting, and that is maybe not the best Association here, but.

B

Kind, so in activity class we have exposed a process called kind, which is this is already exposed, but the property is that we exposed before was just geeta, so there was no setters for it and we run into some scenario about in open telemetry. They have the adapters that they are listening to the to the activity activity. Events like activity start activity stop, and they are listening to the to the activities that creates all the way which is like you will just constructed using the constructor and of course, I mean like.

B

So all the activities doesn't have any kind. That mean, like the kind of property is not there in any way in India, and they wanted I mean the adapter in the oven telemetry. They wanted to set the right kind on the on the creative activity when they they receive the event, so the samplers can can do its job. I mean using this I mean using kind activity, so so that change here is just exposing the sitter I mean this is only changing. This part.

F

Can they do the same thing by adding more attributes, or something like that? It seems like this is a bad compat. Yes, other ways to do it seems like changing kind in the middle of activity. Processing is slightly dangerous operation.

F

Like what, if the first sampler decides to change activity kind to something else, and then the second sampler season, Evalia like we were very careful by allowing changing the name, would like, we I think we even added a method instead of making the properties edible so being.

A

Able to change.

F

Kind whenever might lead to her to debug issues when one part of the system changes, kind and the other part of the system doesn't see it. So if their only motivation is these become person errors? Can we figure them out through other means.

B

Don't you think about, like you know if you use this data and you want to somehow to send it to some other party, which is not it's I mean like some customized pieces, which is expect like to have this right kind on the activity. I'm, not sure this is having me like some attribute for something like this will work in that scenario, but yeah I understand your point. What what how this would be dangerous I mean to change the kind in the middle I mean like for for other activity. I mean.

F

The same way you are enabling seems to work by changing the kind by switching from internal to something meaningful, switching.

A

F

From something meaningful, meaningful to something, less meaningful would break other consumers like if you have Forks two of them would prefer internal, and two of them would prefer kind client so which one which one to set an adapter, the one that would break the first or the one that would break. Second.

B

Mainly I mean like listening to the client, HDTV client or is v-net and because II have agree about, like you know what kind of what kind of activities that is coming such components.

A

B

They have, they have a like. You know, HDTV client, it will be like a client kind. I mean like so there is no confusion about like sitting a different, different values here, but I'm not sure about, like any other scenarios, that people can change that, while while they can break some other parts, yeah.

F

Try to keep activities like less mutable, past creation, because then.

F

My sellers can attract to stuff, and things like that right. So, if like if this scenario is only motivated by supporting all activities, maybe we should just fix those or me move this up right, move this information. We are attack or an attribute like detected kind. That exporters would know how to consume. Just not make one of the major properties on activity Mirabal just to fix that.

G

B

Me, let me try to ping, say: I mean like the.

H

B

Who was using the implementing that doctors here so I.

I

B

Many gives some more feedback about that. Just one second,.

B

Okay, maybe you can move on, I mean like leaves of spending and Pepsi's you're joined I mean we can. What can we discuss hanging.

A

Yeah I wrote it down, I mean I'd, really agree with Pavel like it. It seems unfortunate to change a major invariant of this only for backwards-compatible to call this fit.

A

But, on the other hand, I have to admit I don't see like well, we would do differently without like making it super complicated. Now yeah.

E

I mean: is it how hard is it to clone an activity into a new activity with just the kind being different?

E

If it's easy, then I would say the activity should not let this change and if somebody has a complicated civilization.

A

Yeah I mean the question also is like: where would you do this? Like you said you have to assign the property by man with that assignment happen,.

B

A

When would you so.

B

Just listener to all the activities so when an actor tickets started or stopped so the adapter will receive this event, and at that time the adapter will just set the right kind on the activity. At that time and later the SDK I mean the ultimate realistic, a will receive this activity and it will send it like to the sampler. For example, I mean the sampler could can use this kind and the activity to this to decide to do something with it. So this is how torx mean like the halogen are.

F

We planning to enable this adapter on doublet framework and older versions, of.net core.

F

Supposedly the new doublet core, we can just produce less activities for everyone to consume right.

B

What do you mean? I mean like what you will do is with the HTTP client, for example, and it would produce activity with the correct kind, because we know the current property are concerned, but but I don't think we can change that now, because this is going to break all the listeners right.

F

But you are saying that old activities didn't even have the kind, probably so nobody was reading it before so setting it now is time I'm.

B

Not talking about the kind I'm talking about the app compat in general about, like you know, if we change how we produce the activity on the HTTP client today, I mean they. It would breaks all the listeners because the listeners will not get this event anymore. We can practice ball. Okay, we.

F

Know how to distinguish between subscriptions.

B

Yeah, but do we have a time to do that now? I mean like just to change the whole HTTP client that needs video up net. To do that, we would eventually have to.

F

Rely on adapters to fix things up in this activity right, we would do that work somewhere else and then only open telemetry would be able to consume the right information and everyone else would get internals forever, but I think if we scroll back to just kind setting the correct kind on activities is not a break and changed because the property has never existed. So it's really against. We can set it even on the existing. It should be client or exponet activities. It.

A

Also seems to me that if you like, it sounds like what you're proposing Tarek is that the that the listener is changing its assigning. The kind right.

B

Not the listener I mean whoever creates the activity is the one which is what she's decided about it. So the way it works about that, if you want to create a new activities in the memorial, so you called activity, source, dot, start activity and one of the parameters you pass is is the kind so when you, when we creates activity, so we will create it with the with the right kinds that passed to us and later anybody can use this kind of activity. I mean kind property to do whatever you want.

A

Sure like, but instead understand like okay, and if that's the case, then why would you ever have to set it after the fact.

B

To support the old way or house activities get created before I mean like if somebody created activity using the constructor, which is they don't have any chance of any way to save the kind at all. So if somebody got any activity created all the way see, they always have the kind set to internal. So.

E

Does that mean that you should just make the constructor internal and there will be a breaking change when they upgrade and they'll fix to the correct code? And you didn't need to make this piece edible. I.

B

Don't think we can make this could be some other some other scenarios that maybe it makes.

A

B

To use the constructor so one that one example of it I mean like like today, I mean like why an activity source. Why we start activity we just create and start activity was in the same time. So if you want to just to create activity and not starting it, so the constructor would be the way to do that.

E

Okay, so a construct I mean like really. This is the like. Maybe it would make sense as an inert e if that compiler feature was in a supported language version, but it's not and so like, and it only might make sense, but really it seems like you're, either missing a constructor parameter or the factory method. That is currently the only thing setting. It should be the only way to build it, but the you have two ways of building something and they don't. They have a disparity.

E

So if the kind is a required property, then you should adjust the constructor to make it a required property and yeah it'll be a breaking change, but it's a breaking change from a preview to a stable and that's better than making a thing in the API that you don't want, because API last forever and bugs last for a release.

F

The other thing is, it seems like this is a tiny part of the conversation of bringing activities we produce into the new ecosystem we created for the new listeners and the open, telemetry stuff so like this might fix one issue that we found, but how many issues haven't we found and I would? Rather it get some strategy around dead, then start fixing this corner cases that we discover one by one, yeah.

A

Also, like I mean to be clear, activity exists since don't accord all right, so that so does the constructor right, but I mean there's no reason why we should, but we couldn't add a new constructor. Overloaded takes an activity kind in, for example right. So that seems like like if people use constructors today, that seems less problematic to me to tell them. You had to use the overload to pass on the activity kind than making the kind settable on the on the activity, because.

C

A

Been in it only in the constructor argument, it's just syntactic difference right. It should not matter for the call side, yeah.

E

So activity here is well activity. Context has never shift activity is new I'm. Sorry.

E

Starting at the wrong stage.

B

You can ask them about what they think so see you Riley I mean like we, we are discussing about exposing the activity, good kind of set up property, property, property and I. Think Bob will have saying that, like maybe we can just use some attribute or something to work around this, because this is needed only for up compact I mean for all the compact activity, and it's not really useful for for the.

A

B

Also, the it could be like dangerous I need to to change the current activity after it get created.

A

B

What do you think about that?.

A

So before you talk quick heads up, this meeting is streamed to YouTube. So just keep this in mind. If you want to change your answer, based on that.

J

Yeah I understand so may understand. What's the alternative approach like having additional layer similar like Howie, how we have the like I remember, we have some string information in the activity as well, where we can update the name and- and we introduced think of this play name something like that. So this the same thing works going to add.

B

No, I mean so I 32 the kind new thing exists in activity before but first time when we supposed it, which was a.

A

B

The only way to have the right kind, ons activity is why, during you, what why you are creating the activity itself, the only chance you set the car yeah, so the alternative I mean, like you, know, one of the bubbles I just use some like kind of attribute or tags on the activity to say to save the kind and later you can consume this tag. Instead, yep.

E

I mean two things that seem plausible to me would be either from the the bucket of tags. You figure out if it has the default value, which would need to be something different than something it can be explicitly set to, but if it has the default value, then you'll go find it based on a tag and maybe remember it or not. I, don't know if your tags are mutable. The other would be that whatever the default value is that the setter only works.

E

Well, it has the default value, but once it's given an explicit value, the setter then throws which would be effectively a net. Only property I.

J

See in this way, it sounds like we're trying to having like have this mechanism just for the internal work round, and it's not something we'd like to expose to external customers, or at least telling them like this is a way they should use on the other side by making this a public like a decide method. It's it's more like we're we're giving the recommendation for customers if you have legacy activity- and you also need this feature. This is the right way for you to do so.

J

For you to walk around so I, wonder like if the intention is, we want to scope this down only for our like existing, like ESP donut or the legacy systems, or we believe, there's a potential. These folks outside might.

F

Do we plan to use this mechanism to support dotnet framework or dotnet core 2 or 3, where we didn't have new activity classes and we're not producing new style of activities.

H

We already support that, like we already support collecting a CDP client base, p.net asp.net code from older versions, right.

F

But you were planning to use this feature to keep supporting those and change activity kind for them. Yeah.

H

We do we have alternative, like I, don't think there is any other alternative for us to support the already shipped versions of these libraries.

F

Ok, I think we like for a new customer I, don't think we should advertise this feature for new customers and users internally to enable collecting from old I spinel watche framework and then for custom for the new customers just tell them to use new style of activity and where they can start kind. The way we internet it visit.

H

Yeah, so that makes perfect sense. This is only an issue for like all those libraries which are instrumental using the old way for any anyone who is onboarding through the activities or slash activity, then they would be setting it in the intended way, like you said, but we asked for this to be suitable only for handling the legacy scenarios, but if there is a way around it, then I'm, fine with that, but I'm not yet clear. What is that way around so.

F

Would ending tag or an attribute to a legacy activity that would mark the preferred kind that we detected work or my missing some scenario that, where that doesn't work.

J

Yeah, it will work for for all the Microsoft's internal like stuff for customers. There might be one scenario so I'm.

J

Yeah, so if, if there's a third-party customer, they have the legacy, libraries that is instrument eating the old activity by making this gut set well, basically giving them the message that they can use this method to to move forward. We use a something like a custom property or a hidden tag, basically retain the customers. This is not the official supported path and you have to renew your library.

J

H

I'm fine with that, so essentially, instead of using the sector, we store the kind as a special tag which we don't really publicize.

J

Yeah yeah, so this is more for Microsoft. Only not here.

F

We we are not breaking because they didn't have can't properly before and they're, not getting it now, but we're not we're not taking pictures from there right. Yeah.

H

Yeah, that's right, so you.

F

Have to use new activity, okay system infrastructure, whatever.

H

Yes, that part I understood. So my question was how about exporters like because exporters expect the kind to be coming from activity dot kind. They don't really know how the activity was generated. So we use some special methods like activity, dot, get kind and internally we figure it out either from tags or the actual kind itself.

A

Well, I would expect you to change the setter. Sorry, the getter in activity kind right to just say like. Let me look for the tag at the tag is present. Then I return whatever that thing is otherwise he'll return. The actually.

H

Stored kind, okay- and this is something which we have to end in right, not not from the activity class itself like we handle it in open, telemetry, no wrong way. No.

A

H

A

Suggestion would be to actually change it in activity itself, so that somebody who actually just curious for the kind that would that would get the updated kind but then effectively.

F

Give disastrous error: that's I,.

A

Mean it depends on.

F

What you wanna do with.

A

It like that I guess, if you want to transparent and that's what you would have to do right, you would just basically less advertise the back to all right.

B

So we could like I'm, not necessarily recommending that, but we could in internally I mean the implementation. We will expose the setter as I'm proposing it, but we allow the setting the setter will be allowed only if you created all the way activities, but if you created it, then it was a new way. We would not love that well.

B

This would be a very confusing and like when I mean.

E

That's effectively a one-time call setter right.

D

E

Okay, I think making a simian it only property is reasonable. Essentially what you want in the property setter is in addition to setting the field. It sets another boolean saying whether the value is real or not, and if it was set via the center, which, assuming it's either, whether it's public or internal. If, once the values been set via the setter, you don't search for the tag anymore, but if the setter had never been called, then you'll search for the tag.

B

But the only concern is like the consumer of the activity housing the sitter is called before or not like. This is a question.

A

I still struggle of this notion of like existing customers, because I mean if we, if you have an existing customer, we asked him to call the setter. Then why? Wouldn't that existing customer who can call the setter also construct the activity in a different way? That's the thing I just don't understand like we asked them to change code, regardless right, I,.

E

A

It's that consumer is the.

E

One who wants it set and the producer is the one who needs to set it.

H

Yeah consumer is like like open telemetry exporter, which exports.

A

To some back-end right, but that that's kind of what I said earlier to tarrick, that sounds to me like somebody who consumes an activity, is now mutating the instance effectively. Just there two different data, and that seems problematic, yeah.

E

I mean: would this be better solved if activity kind right now, it's zero value is internal which has meaning if we inserted a different zero value which is unknown, and then it would be up to the consumer that if activity kind comes back as unknown, they need to now figure out what to do about that. Like you're telling them no activity kind was provided.

H

Yeah I need to check late because the spec for open elementary says the default should be internal. There is nothing like on yonder.

E

But this activity class apparently predates our notion of integrating with open telemetry. So we have to deal with the fact that we have a thing that predates open, telemetry the create from activity if it has a defaulted parameter that can certainly default to internal, but this type doesn't represent an open telemetry activity. It's it represents old, stuff, plus open limit reactivity. Now you have that problem.

E

I mean you could do. Another boolean, which is activity kind, is known, but that seems weird.

B

E

There's an easy answer: activity kind: it becomes knowable of activity kind and it defaults the null.

E

It doesn't have one or it does and then now you don't have to deal with the fact that you've changed the numbers from the spec.

A

Yeah I'm not sure that makes it better though, but yeah well.

E

It makes it clear to the consumer that they need to handle the case where they are looking at an activity that, for one reason or another, didn't have a kind yeah.

F

Quick question I know we have to style of diagnostic source methods, some of them produce activities, and some we just get a callback Kim. Instead of trying to adopt an existing activity that old code produced, can we subscribe to the callback that doesn't create an activity and create a new style of activity on our own?

F

I

F

Avoid the entire trying to modify the old activity story.

B

Yeah the concern is I mean we explore that, and the concern is that locations remember. This will require us to our requires open telemetry to create an activity for every callback.

F

Yes, but we are getting rid of an old activity, so it's the same amount of allocations right. There are ways to subscribe to diagnostic sources that do not let them create activities.

H

I'm not aware of that way, because we only listen to Diagnostics over like on or off activities like not really part of the fallback. It's taking something like implicit, which is getting.

H

It's always created new substrate using diagnostic results.

B

So yeah, so my understanding is like, like HDTV client or is being that is creating the activity. Anyone right, if.

H

There is a subscriber, yes, yeah, I mean like yes, that's what my understanding is for, so there is no way to prevent them from creating the activity we only have to like ignore it and we create a new one, which we tried and gave up because of email occasion. So.

A

Given them, you know, 45 minutes.

H

A

Like I suggest, maybe we should take this particular issue offline and we consider the design options we have here, maybe with a smaller group, because I think we probably wanna make some progress on the other proposals here at least see you know whether there's more issues or whether we generally like the rest.

B

Group decided to take it or leave it.

A

Yeah I mean like yeah I guess. The question is like: how do you respond to pebbles criticism that it seems problematic to have a core concept like kind mutable after the fact in general, right and I think that's probably very fair criticism, yeah.

E

I would recommend the design that is not openly mutable. So if you need this public setter, then it would need to understand you've already started. The active you've already done things with the activity it shouldn't allow this to change anymore whatever that means, and that would be my recommendation.

B

H

So the general agreement is, we don't want the setter, but, like a smaller group, can decide like how we work around that limitation right yeah.

A

I think the I think the the I mean having a setter the base. It blows up with like too late to change exception. Kind of thing right like that seems unfortunate, but it is the thing: that's workable, I think the primary concern from proble is that you have an activity flowing through the system and then all of the sudden, very late in the game. Somebody decides to change the kind, but now you have like half the information with the old kind, half info of the new kind right. That seems bad good.

A

Alright, so then, let's go on with activity attributes. Yes,.

B

So today, Authority have some things called tags, which is a key value, bears from string to string and in open telemetry. They have exactly the same concept it's called attribute, but the difference is they can map from string key to any kind of object. All the different kind of options like it can be in numerix. It can be poly and can be arranged. It can be. Something else is not it. It's not necessarily to be string daddy. So the proposal here is like we are exposing a new way.

B

I mean to set this attributes to be from key to object instead of key to string in some time we are keeping the compatibility. So we are not. We are not changing the behavior of the tags today, which is string to string, and you can think about that abuse is like a superset of the text now.

B

So the proposal here is to have a new, a new method called set attribute, which is you can set the key and the value, and we will have a properties called attributes which is, can return use ienumerable. This should be a memorable, not a numerator of key value, where of string and all its.

F

What would we tell customers? What should they use, set attribute or set tag for ID tag? Yeah.

B

Moving moving forward, I would say we recommend attribute and as it would be, the superset of of the text. So the.

F

Would attribute enumerator return, the text to yeah.

B

Yeah, as I mentioned you to be so versatile detect so.

F

B

Said sits a cold, add tag, it will get add to the tributes it will be just one list, but why me returning the tags with the strings if who returns that abuse.

F

We will return the can, we fix that, add an add tag, overload and another way to enumerate them. That would also return objects, I.

B

Don't think so because it's returned I'm off string, key parameter of string and string.

F

But imagine they add, like tags with objects, properties that would return string to object like I done. What I'll tell you might worry. My worry is that, like basically tags is one of the most useful things you do with activity. You attach nice information to it and make it available to all the consumers.

F

Now we are working our main API and considering our convoluted absolution absolution story, it's slightly scary and considering main thing that customers use is the bad method and we can fix that by adding another overload and we can figure out how to do an iteration in the way that it works. So I'm slightly worried about 14 everything just to get object, boilers.

A

Yeah, basically, you now have three concepts, but you have baggage tags and attributes right and they differ very subtlety like like the difference.

C

Is very subtle.

A

Your I agree with that. My problem is I. Don't think there is an easy way to avoid that, because you know adding an overload for air tag. The tags object would be fine, but then the question is: what do you do on the outside, and so you would at least have something else like tags, tags, typed or something right or you know, tag objects or something that is. You know parallel to the text that returns, 91 will have key value pair right now and so.

C

A

But at least have to tell customers used other thing, because otherwise you're suddenly to stringing the objects into strings right and that may.

F

Be fine right, I would say that is still better than yeah. I have another concept just.

B

As the other other thing, I want to point out, if we have overload like attacks, approaches, take object, so open telemetry, specs define the way to add new attribute or the new tags. It's not the same ways that we used to add the tags, for example, about, like you know, they are saying that if you, if you add, if you add a new attribute which is with with with the keys that it's already added before so you update the vendor, not adding a new one.

B

So this means that the new EBI or the new girl load will have a different behavior than the old overload.

F

We have, we don't have to implement that part right like this is a part that we're.

B

Considering so what what do you suggest for people to do them? I mean they have to be myself right to do? What for people.

F

B

Ensure the behavior I.

F

Mean we can add, set tag, method and add tag overload: I'm, okay, adding, like helpers around text concepts, what I don't what I kind of want to avoid is having another concept like it took me two years to remember: what's the difference between tags and packages, yeah.

A

I think that's yeah, that's the thing. I agree with tip the trouble is and I think that's something we probably need to look into is like there's. Also this problem of this hot pregnant state right away. You kind of merge the concepts, but then you get broken by subtleties that you didn't expect so, for example, would be very bad if add tag would now suddenly also override and not just add and throw an exception at the key already exists. So in that case you would have to at least add settings.

A

You know sad tag and add tag overloads right, which is fine, but then the question is what happens when you like mix and match now of two string right, so we would have to make sure that the behavior that we get all these api's are at least self consistent and then get away with the fact saying that, yes, things are called tags not attribute for activity, but which seems fine, but that.

B

A

The verb would have to do right.

B

Tags anyway, I mean like ZZ using attribute names right, but.

C

B

Is why I decided to just Atul to have this name so will be consistent all over.

A

I mean like well the main, but the problem is like you're not entirely consistent, either right because the quit yeah. Basically, the question now is, if I'm, using this API by itself, right, like I, am NOT an open telemetry person I'm, just the person using.

C

A

Trying to make sense of this, you present to me now baggage tags and attributes how the whole.

G

Line magically.

A

Know that what I really want to use these attributes.

B

Yeah I mean from the mr. signature. You can. You can understand that I.

C

Mean searching.

B

Either a string string or you're setting a string object right, I mean.

B

Between tags and baggage, I mean like have to decide to use text you have to. You, have to go back and read some mutation or something on the message to understand the difference. Yes, this would be the same case. Yes,.

A

So don't get me wrong, I'm, not trying to say, like you know, attributes are the worst that you ever write but like I, think they I think the the the general problem that I think we need to I think somehow think about is that these api's are already complicated right.

A

That's the thing that we have received this feedback multiple times and it seems like this is not necessarily helping this, but it adds one more concept to a type that already is not necessarily super easy to use, and then the question really becomes in the long run like. Is this helping or hurting right?

A

Because you're also like, as you said, you already say, attributes is a super set of tags, so you basically have to round trip those within that as well and now the question is: is this is: is the result understandable right or are we? Are we trying like I'll, be adding too many concepts that this type is not the super convoluted thing where it neither works as ABC outside nor as an open, telemetry equivalent right? That's kind of my concern that we did we bolt on too many things where it becomes hard to understand and I'm.

C

Not sure a tribute.

A

To something that kicks it over but like it's, it's really an example of something where we. The only reason we add. This is because the AI ship that we have is slightly different from what open, telemetry ones.

B

B

Yeah from my point of view about you know we're having those because activity today, it doesn't support that this is only.

A

Yeah yeah, but you, but if you rephrase it like this, but if you say our problem is the tags are only strings, then I think the solution to that would, and you know the other thing, is you only support ad, not sad, I, think in that sense the the fix for that would be. We wouldn't introduce a new concept. We would just add overloads to the existing concept right. The only reason why we have now attributes is because open telemetry happen to call it attributes as well all right. Yes, so that's.

B

Another thing I.

A

Think Pavel is right like if the, if the two things you want to fix, is accept, object as well and allow setting, we will just add more overloads.

B

Okay and and the property to return the list, how we were going to do that yeah.

A

So this money would have to call differently right. We would have to. It would either have to say that tags only returns, the things that are string x or we would have to say tax returns, the objects two-stringed.

A

So that's something that you would have to decide. What that? What better is what the better outcome is, and we would probably introduce a new property and then obsolete the old one and have people like use the other one instead, but.

B

F

Few more concepts like it's one, less concept in the area where we have we already have too many and like it's and one more, is 50% more and, as FEMA said, this step is hard to use already so adding like knowing at least between the tags and baggage. There was a real difference. One and you better understand it before you use the type right or you'll waste, all your network bandwidth transferring stuff around, but between tags and attributes.

F

The difference is so minor and finicky and tricky that it's not worth having a concept for that and considering most people would only create instances of activity and set values and only we're very rare. We rarely consume them. We can even create a method in numerous tags and explorers would use it and that's fine, but 90% of our customer base, who add tags wouldn't be affected at all. I.

A

Mean the other questions like how common are tags being used for they do if they're one of the most popular things on this type, you're basically swimming against the mainstream, but if it's very rarely used and I, don't think it matters. You could just do one of the others. We can either do what you know Pablo suggests so that we can literally just obsolete the two properties.

A

Sorry, the property and the method, and then hide them and say exhibit II never had but like the volume of hiding is that hiding only really works if the thing is really used. If the thing is super popular, then the problem isn't the fact that it's an intelligence. The problem is that everybody has to update their code base.

A

But if you don't obsolete it, then the problem is, you do basically permanently present people with theirs tags and there's attributes and I. Think that's and then you would tell people like. Well, you probably don't want to use tags ever. You probably only want to use attributes, and then we are basically back to square one where, in order to really gives us guidance across you kind of have to obsolete the thing.

B

Okay, so so what is like to have overloads for attacks, for example, and and to have a new property torture? Yes,.

A

Yeah I think what I would probably do is I would I would do I would address, I would add it and I would basically add an overload to add tag that takes object. I would add, a new method called set tag that it gives you a key and object and I would add a new property called I, don't know, tag objects or something that gives you back an ienumerable of key value, pair of string, comma object and then obsolete the one that will trans you just the strings.

F

How frequent is the snare where we need to write the tag? Can we get away with having a bull override parameter, so we don't clutter the bear surface with more stuff yeah.

A

I mean in that case, I would probably literally go with new method because they already are set methods on activity anyway. So we already have.

C

A

Id format set parent ID, and it seems like set tag, would be more consistent than an ad that takes an overload that takes an override property. Sorry, a parameter, because it's also not what we usually do in the VCL right. We don't really have ADD with override semantics.

B

Okay, so we'll have attack, Overlord will have sit there and we'll have a new property like tags right.

A

You can you can pretend that the s in tech stands for Strings. We had a new property called tech. Oh no.

C

A

Think you probably needs like tag objects or something.

H

H

Mean I don't have objection as long as we have a way to set tag and retrieve all the tags set. If so, so, this is very much. What do you, what we originally proposed? We just changed it to attribute somewhere in the middle, so we were okay with the original 1s. Also, okay, fine, sorry Riley! You were saying something: yeah.

J

One minor question, so the add semantics allows you to have duplicated keys right, so you can add, like foo, equals to 1 and let her for equals to a string, hello world and with this like for adding the object, do we expect that to have the same behavior or basically, we will overwrite the existing one if the key does exist, no.

B

We have, we have two misses, one is called attacks and one is called C text so that exit is going to add anyway I mean it will not overwrite anything but 4 6 4 6 is going to overwrite. The key I mean the value of the key if the kids existing before so we have.

J

2 meses, not just one, and we provide the overload which accepts the object. Do we also have the add tag which takes the object or with only a lot, setting no.

B

J

C

J

Yeah no more question from outside the works for for me for open time, sure I think if it works for open telemetry, so we I mean we are good from that. Sir okay.

H

B

Smooth to the next.

B

So the next one is is regarding the as, as we see, I mean like activities supporting tags or attributes and ultimately, as I mentioned, define the ways that you should. You should have that you collection created like it has to be in order. If you sit and you attribute, you have to overwrite the old one and they have also extra thing about like.

B

If you send a lot value, it will remove the entry from the attributes and the attributes get used in other different places like in in length activity link and it's used in activity event so and both activity link and activity event. While you constructed it, you blessings attributes to it. So we are just proposing to have a new type collection type for for zetrov use which is can conform to the. What joven intimate respects is recommending.

B

I think if this is was suggested in the first round of the guys lost previews, and we said that it's not needed during the time, because because consumers can use just lists.

A

B

Right now, I mean after defining more requirements for it. The list will be would be not enough for that, and so the proposal here is just providing a new collection type just to to to be easy for consumers to create the attributes, collection or attributes list.

B

F

Me clarify: why would this matter for dotnet developers like the developers are used to collections working or least working in one way or another like? Why would they want to set something to now to delete an item when what they used to is calling the remove method to delete an item I.

B

Think the same way that it was just suggested to mess with themself I mean they added this for good reason. I believe right.

F

I think it might be like it might be a leftover of some regular implementation like I'm, not sure how dictionaries are implemented in Java, but I suspect that, like adding an item, would now might be the way to remove the key or something like that. So I'm, like as a consumer of the library if I want to I, want to have some final set of attributes right, yeah and I.

F

Don't see why I would prefer using open, telemetry collection, semantics that are weird. My dotnet I versus using a dictionary that I'm using every day and I know how it works. So the dictionary, for example, will.

B

Not keep the order of how you add its attributes, so I think also I think this is lost, Russ lost review about, like you know, maybe there would be some limitation to have how many attributes in the list- and all that would be very important at that time. If you want to have this limitation.

F

F

Then I would use the list if I.

B

Care about order, because.

F

C

B

You want to also about, like you know, settings all value again, I mean you have to do it manually, but the thing is most of.

F

The time when I create this list of activities, it's all done in a single place in code. It's rarely a thing that, like goes through five different layers of system to collect weathers right, so I. What I imagine myself doing is I, have a method. I have couple values, I put them into a list, and that's it I, don't do some crazy, complicated calculations to figure out the values?

F

Okay, no often would you need to remove an item from this list versus just adding stuff.

B

Yeah I don't know the answer for this question, like I imagined, like in telemetry space exhausts for some good.

A

Reason but I don't know. What's three.

B

A

My I mean my question is like: why do we need the collection in general like it seems like the pattern that we have on activity? Is that there are methods, and then it is basically just a way to retrieve the existing items as an idea mobile. So it seems we're getting away with not having the notion.

B

A

A collection to begin with and if.

B

You go down a little bit, I mean you will see after the event and activity link, I mean yeah, so you you construct the test event or activity link and passing this attribute. So you have to get the list of that reviews before creating those. So you have to have a way to create this list.

B

Yeah I mean like if, if we think that like we don't have to be 100% like conforming about like the behavior of the open telemetry, so that's fine, but we have to be explicit about that. I'm.

C

Religious ordering used open, telemetry.

B

Yes and I mention about like one of the scenarios like you could set some limitation on the number of attributes you have in the list and at that time this order be important because, like what's added first, it would be have it would be high purity to stay in the list if you've got the rest.

C

B

C

That's a common need, an open telemetry to cap, the total attribute count and just kind of truncate, something after that. I.

J

I will say, one common need would be checking the actual hi and see. This is something we support or not freedom, how? If the user added some attribute, which cannot be series with many to handle that can we.

F

Do it in constructors of link and event.

J

B

Okay, so you're saying there is not that important I think.

J

The other is important, however, I think the the scenario probably is not the like the most frequent one and I think there might be a more frequent thing is we want to enforce the type like certain types should not exist in that, yet.

B

B

Open telemetry side right, yeah.

J

A

I mean I will say like in general, like I'm, not a fan of super NASA generics like I, enable of key available income object because they're not pretty to construct, so in that sense, having a collection type makes sense, but again agree with parallel that it seems a bit weird to have a collection type that very odd semantics, like usually when you have a dictionary, and you add something with a null value, it doesn't magically delete the key.

A

If I can adjust the order, dictionary or ticket away like this I mean we do have that already right, I mean in the sense like they already are kind of I mean the I mean. My first question is what what semantics do we have? Do we, the kind the same key exists multiple times, because if so, then it doesn't even work as a dictionary right.

A

So if a man is sitting based on what was said earlier, it seems like you can't have the same key multiple times. Is that correct or is.

J

There not quite no, so we open times we, you cannot have the same key more than once in in donut activity. You can have multiple tags that have the same main gotcha.

A

And yeah any reason why we didn't take a dictionary here. Why did we take animal of key value pair because are there any other types people actually use today? For this.

B

B

A

So it'll give you the same order as enumerable right, because well the interface doesn't specify it. So, yes, in principle, you could imagine, there's somebody in front of the dictionary that does that right, I know also doesn't make any guarantees on ordering either it depends on what implementation you pass to it right, if you like, if you make it by an array, then.

G

A

Yeah I mean like that's: that's why I'm Jenny I'm not opposed to having a custom connection type that you know, makes guarantees about order and other stuff, but, like hey I will say to me like calling a method called add with a null value that no results in this key being deleted seems fairly odd. Calling.

C

A

Rafa now video and the item being deleted, maybe, but that one is also more like a reflection, maybe of the fact that null is not a valid value. To begin with right, that's more like all the old dictionary make a hash table worked right, but you can't do between a key having the value null and the key not being in there in the hash table, because either way they betray me they would be now and that's the behavior you want, then that seems fine I mean that's. Why I'm saying for the setter?

A

I can probably get behind that. But if you say ad should remove the thing if the video is now, that would be very weird.

B

Okay, so from this discussion, I am seeing like you know, we don't have to expose this collection and we leave it to the consumers of our reviews to create the list themselves and what whatever rules I mean.

A

I guess I'm trying to say is like to me: the collection would be fine, I wouldn't call it attributes. Probably because you just said we don't wanna have attributes, would people a tax collection or something but like it's.

C

A

You know I think having a collection type wouldn't be bad, I would probably still make it implement I dictionary because you can write and that basic is people generally speaking, the right semantics that they want, and then the only thing you would say is you would I would not have the set method. I would just follow the normal patterns and I would basically just say yeah set with an indexer or whatever.

A

If you set it to an elevator, you just basically delete the key and then the thing you have to basically do I think in order to be consistent as saying if anybody asks for a key that doesn't exist, you're to return now.

B

Use a dictionary for that right.

B

Fine right, I can't stay, I mean I, can still have the unit and and.

A

I mean that's setting of interfaces right, it makes no I mean we already have. You know it is in the non-generic a so much so for the generic case, but we do have implementations of I dictionary, the non-generic one that is order preserving.

A

Okay, so I think that's I forgot about the type is we have this weird type in the VCL that you can either index by key or by index? And so that's I think it's used for someone. The networking stack, probably one for very similar reasons, because Harris I usually also order dependent but like yeah, it's just yeah I mean the questions also like to do the generics buy. You anything can pop. We just get away with I dictionary to begin with, but I think having a custom type that avoids these stupid.

A

Nested generics I think it's probably useful for the API. The only thing I would say is like make sure it did these collection types. You know generally gel with the VCL right there follow normal conventions. You can use them up. You can use collection, initializer syntax for those and then I think it might be more usable than the raw animal you have here, one one.

F

Useful drawer innumerable might be implementing it on existing things. It's kind of a.

F

Little bit of a stretch, but we did this for perf in other places like I- will a desk source. For example, we implemented up on existing objects and reused. So maybe, if you have an object, that kind of already has attributes, but then I numerable is not the best interface to do it at least, would be better. So yeah.

A

I mean like, even even if you think, I knew of key value pairs. The ideal thing for those kind of situations, I would probably still suggest having an overload that would just take the whatever the collection type is called because if somebody is new to this and I just do new activity event and then they see base the item will of key value pair string, comma object. They have no idea how to get one right. I mean yes, some people know they can just new opener in an array.

A

But like it's much nicer, if you see oh there's another overload, it takes a I, don't know a native its attributes collection. Let me just new this guy up right and then you merely have a useful type that you can pass in, but that doesn't mean we have to win to base. What I'm saying is, even if we add the collection type, we don't have to remove the one that Xie movie so how it doesn't overload.

A

B

Okay, yeah I'm, fine, so.

A

I guess: I'm fine with the collection as a proposal, I'm not fine, with the shape yeah well.

B

Well, changes you want.

A

Okay, let me just write down what we just said should be renamed.

A

To something tags.

B

A

That would work for me activity, tags, collection,.

F

Wait: how is it, what was the appropriate name on the activity event and exposes this dictionary? Is it retribution tags.

B

We said that we will have tags the names desi.

F

So what do we need to have constructor parameter? Also name, this tags, I mean productivity event and own activity link.

B

Yeah, we would call it tags yeah, okay, yeah, so yeah I mean like who you stance all the way in our areas.

A

B

Will not use attributes.

A

A

Mean my other question is so what is your expectation for the constructor? So they take these this new collection type and then they copied the contents or what happens with the content.

B

For the collection.

A

Itself well for activity event and activity yeah for activity event. Basically does it does it copy the contents or does it store the attributes collection directly right.

B

We are trying to be performant and we will not do any copying or anything and just keep the reference for what they passed.

A

To us, so is the collection then mute a table or is it in it? Only it's.

B

Immutable yeah.

A

B

But yeah I mean it will be clear about like if you pass us mutable and we change it later. This will be reflected on the activity event or activity name.

B

Yeah, this is the same case for other languages. I mean like we asked it about like Python and Java I mean so they doing the same I mean they don't copy the list or make it read on me.

A

So that would also mean that on activity we would return tag objects then stood as an enumerator of key value pairs, drink on object, or are we already exposing this as activity tax collection as well.

B

No, we will not expose it as collections here, because we want to reduce the confusion why you want to sit and you attacked zero, so people can get confused about like to get the collection and since their attribute 0 is at exit. So we are trying to avoid to have two ways to sit the tags on the activity.

B

Is this clear order? Yeah.

B

So we still I mean like text. Objects will return I in the morning, not not returning collection.

C

A

So the collection is nullable, which is weird.

A

B

A

It's normally not something we do but I guess. If you accept it now, I new mobile I guess you can also accept it now yeah, but.

A

Why do you I mean? Why is it not optional at that point? Well, why do I have to pass it in at all.

A

I

Almost everyone.

A

You want it overload that just doesn't have them to begin with.

B

Yes, simply in the first last time that we said that we need to have overload for the constructor, but I. Don't remember why not ocean, but yeah I mean like if there is no objection, we can make this optional. Prompt I, don't mind.

F

Mommy doesn't have a pair versus a parameter that is optional right. You have another constructor that just takes name and keep the current one is either.

A

That or I mean they could speed weird right now, because you have basically overloads that take their time offset in some. Don't take the item offset so either you go all the way. You just say: okay, let's just have a very simple constructor for name and then we only have one other constructor that takes name optional day time off certain optional attributes.

A

Because otherwise or you just go all the way and just save it, we have all four overloads, which is probably what.

E

I would go with the two.

A

By you mean name and the other one's name.

E

Our name as the only parameter to the simplest constructor yeah.

A

E

Then timestamp and attributes both being defaulted and.

A

E

Assuming that's a timestamp value that you use as the default is actually two multiple. If the other constructor uses daytime offset UTC now, that's not easily expressible via default parameter. Well,.

A

That's why you could say you could make it inaudible, but you could say they don't offset now means default right.

B

Okay, this sounds personal.

A

All right, then, let's go further favorite, you link, I, guess it's the same argument here: yeah.

A

So should I get a triple shot to be optional here and.

A

B

Yeah this last one, okay, so just to give some some idea about like what we're trying to do here. So when you try to create a new activity, you call activity, source, dot, start activity and one of the parameters you pass is the parent context context. It has like twist ID a span ID, whatever stay state and flags and later about like use anybody listening, there's any listener.

B

I mean the listener, will get a call back with these parameters, including the current context, and the listener can decide if we need to create the activity or not, and if decided to create the activity, we will create the activity using this context, which is it includes the trace ID. This is the important thing. So, whatever we pass go to listener, we create the activity and we use the twist Aggies at its best in the enzyme.

B

Here we got a scenario about, like you know, when you don't, when you don't have parent I mean now parent? This means that we are passing the default context, which is doesn't have trace ID or span ID or anything, and at that time the listener will get this callback and we'll have the empty or the 40 context.

B

This I mean like no in most of the cases, this is okay, but we got some cases like in open telemetry. They implementing what is equal probability, it sampler, which is listening to the activity creation and they decide about sampling mmm or out according to the trace, ID part of of the decision. So they want to get a valid all the time, a valid tris ID, while coding them.

B

So this means that they, if we passed that the fourth context, this will not be any helpful for them and they will not be able to do any, take any decision to decide to sample in or out this activity. So the proposal here is to have, in the listener itself, to have some property telling if we need to create a trace ID before calling the listener.

B

In case of we have default context. So if you have default context- and this property is set to true so we'll generate- and you know I'm due for a trace ID for for that and passing to the listener- the reason that we didn't do that by default- and we have this property because, like generating a new twist ID is costly. I mean like it. Have a cost, I mean it's not free. We allow to do allocation and take some time to allocate to to generate like random trendy.

B

So this is why we make we making this property I mean making this behavior as hot and behavior. So somebody want to ensure that getting a valid risk ID all the time. So they have to say this property on the listeners we are using.

B

Is this clear or.

F

B

Okay, so let me submit more, but so open telemetry have this proper, proper, proper property, sampler, okay and this sampler take decision about to sample in or out activity according to the trace ID right.

B

So if we, if we passing to the listener or this sampler, the trace ID, which is invalid one or defaulted one, they will not be able to take the decision to something in or out and the case that we send. The default context today to the sampler is when we have a null parent. I mean there is no parent and we just send the default context which doesn't have the trace, ID or span ID, and this making a problem for the slammer.

B

So we you want to make sure that if the sampler want to get all the time a very traced ID they they want to set this property to true and we insure sending valid risk ID to them and if they decided the assembler, decided to something and creates activity. We are going to create activity using the generated Chris Mundy.

F

But this random Traci the wouldn't affect the samplers decision right. It wouldn't make the right decision anyway, because there is no information it relies on. There is no parent, correct.

B

Know the probability sampler is depending on the trace, ID doesn't matter if it is generated randomly not it just. We have some some way. I mean like to make the decision according to what size it is.

F

Occurring what would the trace ADB? Yes,.

I

F

Okay, so why can't we I mean, wouldn't just using the random class and deciding based on that? We give. The same result means.

B

F

Or just to use whatever random get max int and the probability they have it would give the same. It would produce the same amount of instrumented activities that that are generating the activity. Id would, but unfortunately it would lose the connection between every ID. But is that connection important? Or is that just how the sampler worked before? And it doesn't really matter if we keep the same algorithm as long as we sample the same percent of activities.

H

So traits record for like when we create downstream further activities, they'll share the same trace earliest appearance. So if the sampler just creates a random ID, we thought without taking the actual race ID, then it will result in inconsistent humbling of child activities, because when the child is created, the child, the sampler, which decides whether the child is humbled in or not, it would be looking at the old trace ad of the activity, whereas the parent would how created like some random number so to ensure both are some building together or Samba. Doubt together.

H

We need to use the same, try setting the parent and child. So it's totally fine. If the parent like if the parent sampler just generates a random ID, but then there should be a way to tell the activity creation mechanism that this is idea used. So please use this as the trace ID. Should you choose to create the activity?

H

F

I think there was a at some point. There was an overlord, the sampler that returned additional information. Did we remove like I, think a return to attributes, or example, result with attributes.

H

Can we use that, like in open telemetry, probably I need to check whether that is still part of the spec yeah yeah.

F

H

Seeing that yeah that.

F

Seems like, like a place to put information like this, maybe.

H

Not sure how would I solve that in application insights like application sales, have like a binary result like somebody or out there is no other place to put attributes I.

F

Mean I understand, I, definitely understand your use case I'm just again worried about making the, but the entire goal of this endeavor was to make subscription and production API simpler.

F

If we start adding all the knobs and features to enable particular use cases, probably gonna make it harder or is this feature important enough, then it's lower than that.

H

Are you concerned that this is like we're adding too many like API changes, because I'm guessing these are things which, like library of this, would never change.

F

It's more than it's like a super super targeted, API change to enable these particular implementation of sampler to work like would anything else, use this parameter, or is this just for a probability simpler that bases its problem.

H

That's only use case which I could think of like both application sites and open elementary. They both depend on the ID to make the sampling decisions.

F

Yeah this is this, is kind of I'm worried that it's not a general purpose if I, this is a way to work around the difficulty with implementing this particular simple, simpler and maybe made this versus, but I wonder if there are other ways to achieve it.

H

Is there any proposal like to get the same.

F

Okay, without it it's a lot of implement. There is a lot of implementation detail, but what's all right, can you remind us direct? What's our sampling interface like how what do we take and what do we return and when we ask Center for a weather activity should be accredited or not, I think.

B

We send the whole information about like the context, the length they're in their.

H

Activity creation options, yeah.

F

Mister and the pair listing somewhere, we can take a look.

B

B

Pasted something I think it's pointing to the to the sampling.

F

But I mean our API to activity source in that how? How does that look? Okay,.

A

C

A

Somewhat at loss here to where it's like, and where are we going with this, like I, mean on the one inside and similar property, is not bad. On the other hand like given how much time we spent talking about it like it's clearly, not something that people necessarily will quickly understand what it does.

A

But then again it's a single property, so it's also not like the world is ending.

B

Like I mean part of the suggested names, I mean like generate by default context from like it's a clear above, like what it means that.

A

Yeah, unless there's a mean the normally D, we kind of knobs that we have in DB CL is naming and place where we put stuff. So if something is more advanced, be trying to hide it by putting it somewhere else, if we can, or by making the naming at least somewhat clear that this is something you need to read some box on in this guy. In this case,.

C

A

There's an obvious way, but we can well. You know where we can put the API somewhere else.

B

This property other place, we can sit in there and as I mention about like you know, if nobody cares about the scenario at all, they don't have to do anything, but if somebody needs some specific scenarios that we are trying to enable here, I mean they need to learn about that yeah.

A

So that I think the only problem that is is like, if somebody would have had this issue, would they know that this is a property they need to set?

A

And then one I don't have an opinion on, because the next breath.

B

Yeah I, don't think it would be super clear, I mean they won't jump to it to use it and do it I mean if I mean like mappings a scenario to this property. It's not easy. I mean I, see that, but but I know I exist. Any other piece of me.

A

Yeah I mean given it enough: no better option here like probable. How strongly do you feel about this being problematic, otherwise, I would just say: let's just.

F

um Considering I just took a look at the activity listener interface, it's complicated. You know already.

A

The damage has been done is that the argument.

F

Other option is currently we returned an enum, the sampling callback like if we decided to return this drug. That has multiple things. Maybe we can put it there, but that's even bigger change. So, yes, I, don't have a better alternative, but should do is route ID exposed in any other property name like what is the actual property name on activity context that gets populated I, don't think we have this. This property names should be should at least corresponds to what we can assert yeah.

B

This is why I mean we have suggesting some other names in the comment, so maybe generate trace ID for route context. This is the most one I like. Actually because it's it's it's telling the purpose of this property. You, like you, need to generate the test ID for root, complex.

F

What's the type of activity trace, ID, it's activists residing I mean what's inside of it, it's just a string wrapped in a struct yeah.

B

F

Right reading string: yes, what if it can wrap something? That's not a string and doesn't require an allocation, and we can so how.

F

So, let's say it wraps a battery somewhere in the pooled space. And we point to the byte array before calling the sampler and if the sampler decides to.

F

I mean make it lazy materialize it when somebody else, it's the same amount of allocations right. If somebody, if someone needs to look at the string, it has to be string, but if we can, for example, if the sampler can look at the activity ID as byte, that's a way to make it faster. So what I'm saying is? Can we make so that uninitialized activity still has some.

F

Implicit value, that's random, I.

B

Mean like it's not just one random thing: I'm, you know generate once it can be generated model.

F

Right but you don't have to allocate to generate the random value, how.

B

A memory or something exists or.

F

You generate a long right or too long and store them as value properties and.

B

Every time you want and you random one, you generate a new one. You.

F

Call the segment yeah, you called the same, create random, ok, so just.

B

F

Who doesn't allocate when you generated, but you can create a random instances of goods I'm talking about the same thing like good, good strat is what.

B

I mean okay, yeah I mean I'm just trying to understand. If we go with that, I mean so. What's your suggestion is to to generate this random all the time and don't have this property or what yeah.

F

That does that's what I would suggest always pass, something that looks correct and then, if.

B

Has a performance concern I, hear that multiple times so I mean when you have a default context, which it will be a performance issue, I mean the whole the whole the whole issue. Providing this property is to avoid performance penalty I mean.

F

If it's about random being slow Danya but yeah.

B

I mean yeah, I mean we. We know that I mean like no really measure of that I mean and he's trying to look at enhancing how we generate the random stuff. But, okay.

F

Yeah yeah, if, if it's to prevent random from being golden yeah, I'm, not sure I can offer anything.

B

Else, so what she named, you prefer now like I, prefer generate less ID for route context, but I, don't I, don't know what puzzles me like as a name suggest and Union or social status,.

A

So one name they were suggested in share was all generate route context, race, Eddie,.

B

A

Mean it's long, but I mean for an advanced video.

B

I think that's not bad. Okay, I'm, fine, I.

A

Mean at least it seems like the closest to what you said in text.

B

Okay, yeah I think you think that's all for this review. So the only open issue is the activity commentator I mean: will you will take it offline and let's find out what.

A

Seems reasonable, so let me just add a bunch of.

A

All right, so let me just scroll down to the thing to make sure that what I wrote down captures what we talked about.

B

Yeah techmo object; it should return. I enumerable not IM ready. Just in case sorry. No it's my fault, I mean I, know that I wrote it like this yeah.

A

You think I noticed that, but it did not fix it actually.

A

And then missing a semicolon.

A

Actually missing mom moment semicolon.

B

A

So then this is it we have 10 more minutes. Maybe we can improve the other one.

A

Steve, not you on the call by any chance.

B

A

Skip this one I.

I

Have a five-minute one yeah.

A

I think this one, you might also be five minutes of eight actually, because I think we talked about there's an email already yeah I'd appreciate if we can get through this one. So last time what we did was this, so we basically said we have a chibi request options, implementing I dictionary of string, comma object, because I think the concern that Steven had is we don't want to duplicate the properties storage?

A

We would like to have a single storage for these things, but the single storage, one that we have is basically untyped right and so one of the major of Ages that heb request options would add that things are strongly typed by using the generic here and, of course they messed up the signatures, but that's irrelevant, and then basically the idea is that we said this guy's edible, because the expectation is that things like HTTP extension methods we for Jason.

A

They would eventually construct a message, but they would just get the options in and then basically pass it on this way and then the feedback that the networking folks had was. It was three issues. I would even remember one one I, don't remember, I think I had one versus we just missed one of those types. We didn't actually clear.

C

Up in Turkey.

A

So that one is was missing and then you wanted to make this also editor. Browsable. Never, and you didn't want to make this setter settable, so I think the API ship that was proposed and returned to this one. Was this one, this one busy just lists the key I still missed the generic argument here, but that doesn't matter and then basically the idea is that this guy would be at the walls whenever and this guy would be not settable right.

A

So my concern so I'm fine with everything here, except for this one being missing, because I still struggle with the notion of how would this work then for things like HTTP client extension methods, where the whole point of the API is?

A

Is that the consumer doesn't have to give with the message directly, but they still want to control some of the options, and so the idea that we had was that we would take the options in and then deep down and one of the extension methods been a ver cahors methods where the message actually gets constructed. That's where we would basically take these options and put them into the message.

A

So one option we have is, we can say: okay, the LGB request options is an ID canary already, so you can of course, trivially enumerated and copy the values over. The problem is, we also said we want to explicit interface implementation of this to promote the typed API surface, but then we'll make it really ugly for consumers to effectively transfer the keys from one to the other, because you basically have to downcast this guy here to an ideation area of string, comma object, which isn't necessarily super intuitive I.

E

Mean couldn't we solve that with a copy from method on HTTP request options to copy from another options, yeah.

A

I suggested a set items method that takes the options.

A

There's like a copy, constructor yeah copy constructor wouldn't help. You, though, right because you already have the instance but yeah yeah.

C

I mean if you just if you just want to cat Nate I suppose yeah I mean the reason we suggested getting rid of the set is that none of the other collection types in httpclient are suitable right. Now, there's an ELISA initialized behind the scenes for you. So.

A

That I may I agree with.

C

A

But I don't think we think of options as a collection really, but it's just that it's a it's an open-ended struct really right! That's I think that the way to think about this because you wanted typed right, Emme, I.

C

Guess the question is: will this be initialized once and then just reused for tons of requests connive seemed to think that that was probably not the case in the majority of uses, so we figured it would be safer to just not have a set now, and we can add it later if the if people asked for it so.

A

Maybe I'm missing it, but let it so busy you're saying you you will the velocity of things would not have an options instance and that that's that you can still do that right, because lazy just means first time. Somebody calls together unit you you you initiated right, so you can still have a setter for that right.

C

Yeah we could we just we didn't see a super immediate need. I also I understand that you're thinking about extension methods, one thing that we we want to tackle in probably the dotnet 6 timeframe is: we have users who want to set like per request headers without, whilst still using convenience methods, and things like that.

C

So the thought was that we could tackle all of that in one and kind of come up with a model for how to pass all of these requests specific things in a way that we don't have tons and tons of extra methods. Every time we want to add something new I.

A

Thought that this design- because we thought that is the way we would do- that no I.

C

A

C

A

I think so, I think that's time we talked about this Stephen brought this up and said. Basically, we already have a way to do generic things with the request and I think originally I forgot what they actually I am if they actually ain't here right proposal. One proposal one had this very specific instance of the handler. We didn't like that. One then proposal two was basically more generalized, so we say: okay, we have these request options that we can now pass into all these methods. That do anything.

A

Let me say we don't like that either because it seems now every time we do anything. We had these ginormous numbers of overloads and then we said well, you know. Maybe we just put them on the message and if we put them on the message, then well then the basically the parties that care about this can take them in the parties that don't care that don't care, and then this way it's similar to, let's say our extension methods for Jason's utilization, where we have one overall, it doesn't take the Jason options and one that does.

A

If the default options work forever for everyone, then they don't have to pass them in. It will be somewhere here. If you actually need to pass in per request thing, is you do and if you don't, then you don't right and but like basically, this one is open-ended. We have strongly typed piece and then you means you can almost pass any data. You want to the request. I mean.

E

I can say that we do have the general guideline in the book of guidelines that collection properties should be get only right.

I

E

Should be should return not null and etc, and so that the the general guideline would be more, that we add a copy from or add range or something method that drains one HTTP request options into the one that was already built on the message: yeah.

C

E

Was that because it would be the setters weird, because if you set some properties already, the fact that properties and options are the same thing, so you called like properties, dot, add and then something replaces options where'd, that property go like there's. No, that the things don't mix and- and it's just back to all the stuff we were talking about with telemetry, of like it's fine as long as there's only one writer but right.

F

E

That there's two everything stops making sense.

C

Yeah I mean the initial proposal or counterproposal or whatever it was. We talked about adding first-class methods on HTTP client and this being kind of a first step in the collection that you would pass to those that would be usable by all the different types of handlers. All we're saying is that we would love to add headers to that mix.

C

Somehow it we're just not ready to so we don't want to add methods now, when later on, we would have to add more and if we did add the methods directly on to HTTP client, we could use internals to construct things optimally if those get passed in, so you wouldn't need a setter. In that case, yeah.

A

I guess I'm fine with not having a setter I care about the scenario. More than anything right like it just seems, the setter seems the most obvious way of doing it, but I by the problems of that so yeah. My suggestion would be at the very minimum. We should have a method on HTTP request options that basically takes another HTV request options and just effectively calls the set method in the loop.

A

It doesn't quite do that because it's generic but like that would be logically what ends up happening. So basically, if the, if, if you configure the key and then the the request options, you're taking and also defines the key, what that would override it would not cause an exception, it will just replace the contents. So that's why I would probably call it set items, because the idea would be anyway. Everything you have. That's not in the one that's being passed remains and everything that you know is new will be. You know edit and everything.

A

That's already there will be overridden. That's similar generally see the semantics you want.

C

Pronounced do we think people need to do that sort of copying of I.

I

Don't think so, but like you, don't really know how they're going to end up using this I'm. Sorry, that's.

A

Fine well you've, not that.

I

A

There's mics around so when you say it's not common, like I mean I'm somewhat confused I mean like your bass, is saying. If somebody needs to pass in data per request like wouldn't you by default, not want to replace just add the new options on top of it right.

I

No I mean like it, isn't it doesn't very common to sort of build a pipeline off. Somebody said some values and then you go and like override it and so on so forth. You might want to be able to like do things like. Oh right, I mean in general, because you decide what bodies you want to send.

A

Ose, what you're saying you're basically saying for the most part you I have zero options, and then you just set the five you care about, and that's it there's. No, you know replacement or anything going on yeah yeah no I buy that I. Think to me it's more like overriding everything on the request message. If that's the only way of doing it, the you know to Jeremy's point that might bite us later right business. If the semantics we have it's just will be a place to once.

A

You gave us then, if we ever, if anything ever is present already like at least it doesn't get dropped on the floor and Kim all right. So then, so any objections with like public void, set items, HTTP, request, options, options, I think.

C

My only objection would be if, if we don't see an immediate use for it by the domain, expert and I would prefer to leave it off an ad later.

A

So I mean, as I said, like the immediate need would be. We would want to add it to the HCP extension methods, because that's the Blazer scenario right so like this one was added specifically for blazer support right and we expect people to use these extension methods in the blazer app. So that would have you would have to add overloads to them to take a chibi request options. And then the question is: how do we compose them?

A

With the underlying request message and, as I said, an email you can you just have to downcast the thing from options to I dictionary, ostwinkle object, but that seems ugly.

C

A

A bit a bit Hecky yeah.

C

I mean if we see cases where users would want to like have one options with all of their stuff in it and just constantly copy it to other options, then as I suppose that API could be used. But that seems no.

A

That's not the scenario.

C

Seems a little officially.

A

The the scenario is you call it a method on HP client, but you don't have a request message at all. Right, like the the the koala just says, get as Jason async and passes in a URL and maybe some Jason options right. So another person wants to say: oh I want to configure my whatever the whatever one of those. Let me send. The settings were like I forgot. Like some of these, you know what was it.

I

Saying technically so, Tina so I think we ended up like dropping that entirely. Toe I was like the API, which was extension like adding the HP request options to yet I guess was being dropped.

C

Yeah, so the original design was based fully on extension methods to set properties, but then the other proposal, I think in this API, was to add I think maybe I sent a sink that took the options and we we dropped that, at least from the BCO library.

I

Right but like we do, we keep the one on system, 9sv, Jason, I,.

C

Don't think there is a proposal for Jason in here. Was there there.

I

Was one because, like they get Jason and I think we said like so, the ocean purpose will have that in there.

A

This one, like.

I

A

Yeah, so this is the one where we bases said yeah, so this is my example: effectively Cori right so like this guy, you basic pass in the options and the consumer has no like there's no message in here. The implementation of this method eventually probably constructs a message, but that's basically, the part of a would have to stitch in these request options to have the one on the message.

C

Yeah, so I don't like this, because we're going to in in the future add another overload of this that also takes headers, so I'd, rather avoid having two overloads that you know, people have to use intellisense to figure out what they're doing well.

A

Then let me go back the all the way then like do we need this API at all. Then I forget about this one method like do. We need the request options at all. I.

C

F

C

Options will be used in the future API, so I'm, fine, with keeping this here I just don't know if we need the overloads and then.

I

Like there was also like some pushback from the like a mono team about having like a spindly type dictionary like they wanted something that felt a little more first-class, because we still need the ability to configure these options. Well,.

A

That's what this is right. This is not string lis tired, but this is strongly typed right.

I

No I meant like me, you asked like do. We still need this. So yes, like there is some incentive it. It doesn't complete this jewelry but like if there are plans to like evolve. This API, like I, don't want to like so maybe maybe.

A

Stubborn about it maybe I'm confused and so like. Okay, so let's say we add these api's as they're proposed here right. Let's say we have them and we don't have these method that I'm asking for. So what is the end-to-end scenario that uses these api's in.net five then.

I

um The the browser HP handler basically exposes extension methods that configures these things and then uses them on the consumer and like in.

A

The implementation.

I

A

And what would they look like like I mean they're not listed here right so like? What would that be right?.

I

Selected API is completely new, like the type is in public as yet so I was gonna. Bring that up separately, like I, did help.

A

I guess the reason I'm asking is because I'm still struggling with like how this is being composed, is the idea that whoever consumes this API then end-to-end already has access to the message, because that's kind of what has to happen right. Yes,.

I

It does so it's an HTP hunter on the other side, so there is like a user code on one and the HCP under other. On the other hand, the SG on the pillar currently uses like the dictionary like a TV's property. So it's just a loose bag and we don't want a loose bag because it feels like unstructured I wanted to bring a little more structure to it.

I

So the property foes that gap so and then you, like you, use the the options as a way to sort of make that, like experience, a little nicer and then in five o mean six so I. Guess we improve on that and then figure out how to pass like these options to get from Jason, async and so on. So.

A

Is then the idea that effectively the the party that that deals with this and the part of the deals of this are both infrastructure? So the end user doesn't actually do anything like this? um Probably.

C

Use options right: they would some fetch options if they need them.

A

So yeah, but we're it weird, so I guess I'm still at lost, like the fetch options, are what wrapping this one here or are they actually just this one.

C

Well, presumably, there will be some sort of blazer options: dot, blah blah blah and each of those blah blah blahs is an options key, so you had call request message: stock options, dot set blazer option, stop Baba, ba value, yeah.

A

I get that and then the question is the instance of the request options. Where does this guy go then.

C

It's it's tied to the request message: I.

A

C

You would always have to have a request message.

A

C

Yes, yeah. That's.

A

Kind of where I'm heading, because I mean my understanding, is that for infrastructure code it makes sense because at the very bottom there is a message right so that that's why you always have pretty much access to the message, but the end user code, who actually issues request from the browser they very I mean most of the time you want to stay as high level as you can right. So, like that's why these extension methods on HTTP client exist. So people don't have actually construct a message by hand right.

A

So if we are saying that for the common cases where you want to configure something in the browser, you never have to use the request options because the you know underlying hand. That does the right thing anyway, then that's fine, but if we expect people in the browser who just want to set a blazer option, if we now ask them all know every method you have called so far the constructed messages for you. You can only do that. You have to actually construct the message yourself.

A

Then that seems like a little bit of falling off the cliff right, because you just want to set this one thing and then it's like well now you fall off right. That's why I was asking for this extension method or this this this, basically, where these things are being stitched together, because.

C

It seems like the.

A

End to end here, if it's based on the message that seems fairly low-level for what people want to do, I mean.

C

Needing to construct the message: it's not low-level! It's! It's really how you do anything today that you can't use one of the one of the convenience methods on and often times the convenience methods. Don't do the right thing in terms of the response message like error, handling things like that, so users really end up using it anyway. Okay, I agree with you the this. It would be nice if those could be made more usable but I. Just I, don't want to add convenience methods now and then F to add more overloads to them later until.

A

They feel like it.

C

A

If you're, basically saying that the options people would set like sorry, if you're saying the people that would use these lazy, specific options are already on the advanced path or would be very lucky on the advanced path with the construct messages directly, then that makes perfect sense to me. I just want to avoid the thing where we now have this feature. That is like 90% there, but then you know.

C

Yeah, we wouldn't want every user.

A

To have to take this path right right, so then, in the end, in that case, yeah I'm fine, with this as being proposed without the set items method.

I

Right I mean the extension methods today like live on HP request message, so we.

A

I

Of already expected users to like users that were configuring, it today already were constructing these the messages by themselves. So it's the experience, isn't any worse but, like it'd have been nice to have a hot glue code that one get from Jason. You think yeah.

A

I

Like at the same time like I, don't want to be, like you know, put my foot down and say like no. This needs to be here in five or if there are plans to have all-day pn6. Oh yeah,.

A

I guess that's my point right so that give if I just want to avoid the bishop of broken end-to-end right in five four like if you think there is an end-to-end, but it's the thing that's missing is a is a feature that lives on in the next release, and that's fine but, like you know big, because one of the problems I have with this issue here is that it's on the one inside very specific to vamp assembly. But then we ended up with this super generic API.

A

So I know long I see what the end-to-end is, and it's probably my fault, all right.

C

Then let me just approve this, as is there was one one. Other change that was adding editor browser will never yeah.

I

That must be the.

C

Original properties, I.

I

Don't think um anybody here I mean.

E

We've added EB, never on things in the past and then we get the issue that says: please remove it because it's annoying my tooling in some sort. So it's it's like if we've marked it as obsolete I think that's enough! I don't like we can do EB, never in another release, but let's, let's I, think doing them both in the same one is traits people.

C

So the reason we justify this is none of our built-in types actually use properties, and we don't know if any third-party handlers that that use this either or in general. So so, yes.

F

C

And we think no one is using it at all right now, and it would be a little easier for users if they didn't see both properties and options and have to guess between the two and wait for a warning from obsolete. Attribute I mean.

A

Six percent is not nothing, it's quite a bit actually but yeah.

E

That's that's higher than most of my types. That's.

C

Actually, really interesting, I'm doing what they're using it for so.

A

Yeah I mean like that's, why you know it's moderately highly: it's not moderately highly, you said, make no sense but like it seems like it's high enough that it's concerning I think.

C

A

The end of the world, if we, if it's visible to me honestly like the the editor browsable, is basic just to remove clutter from intellisense. That's really what it's for right, because it will not do it anywhere else. It will still show up in Docs, there was still compile right, but so the biggest mileage you get out of this one here is because the base it tells people they're using the they're doing the wrong thing.

A

They should use the other one instead and so I think that one that one is the thing we should be doing and then a little browsable yeah. The problem is, if you, if you hide it, and people ever have to use it, that's annoying, but I. Think in this case you could invent ourselves a DEP, that's never the case, because every time you passed in properties, you can all pass on options because it really converts. So that should be fine, but like I, also down I, believe Jeremy.

A

Every time we do, what we usually have to you know do something later: yeah I don't have like a fundamental.

E

Objection to it I'm just pointing out every time we every time I can think of that. We've done it, and this could be observation by us. It's big someone has been complained and we've backed off.

C

E

Remember the back offs.

A

Alright then, let's leave it as this and then, if there's an issue we can talk about it.

A

Alrighty then,.

C

A

Was a intense but at least we got somewhere.

A

All right thanks, everyone bye.

C