.Net Foundation .Net Community Standups, 16 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: ASP.NET Community Standup - June 16, 2020 - Performance Infrastructure (David Fowler & Sébasten Ros)

Description

Join members from the ASP.NET teams for our community standup covering great community contributions for ASP.NET, ASP.NET Core, and more.

Community links for this week: https://www.theurlist.com/aspnet-standup-2020-06-16

A

A

That's all of us well now watch and see yep I'm here, Nadia ray looks like we're live somebody on the twitch stream. Let us know if, if you have any problems, can someone else speak? Please can.

B

You hear us yeah I knew him. He.

A

B

A

Some echo from somebody, though, all right cool readings, yep audio everything, good cool ray, it's the asp net community stand-up june 16, hello, Damien and today we're joined by Sebastian and David Fowler, hey.

C

A

Back folks and so you're gonna be showing us some performance infrastructure, stuff I said.

A

C

Is feeding the YouTube audio back in through their microphone yeah.

A

Don't do that look.

D

Me I, don't have YouTube Lupin or.

A

Twitch oblique.

D

A

Things happen: everyone can hear what she needs to hear great, wonderful, oh here we go with the community links for this week. So these are I will publish these in the chat and everything it's a SP nut, stand-up 20, 2006 16. So here's so you know, don't worry, I'll be sharing these all out at the end. Here's what I got for you today. First of all, Jeremy lick nests with a huge three-part series, building a line of business app using blazer webassembly and a lot of energy framework.

A

Goodness so there's there's a ton of good stuff in here. I love this because he's building out you know some real world concerns. So some things I want to point out here. One is he's: got auditing of user account information, so he goes through and he shows how he sets that using any framework.

A

So he's he's, you know, setting up some things with change tracking and then the end result of that is you actually see as people as users are edited or changed, he's tracking that so that's a common line of business scenario that you'll see and so that's cool- to see how he's got that set up he's got some other things with with data annotations, etc, want to point out some other things in the other parts of the series so part two. He shows off some cool stuff here.

A

One thing, let me see he's using some stuff using versioning and he's also got a repository patterning, a pattern here to allow for testing and in the third one and I'm just showing the highlights here. These are definitely you know a couple of cups of coffee. One neat thing here: he's showing is filtering with debounce, and so in this one.

A

This is there's a search interface and the problem that you run into is searching where you allow typing, and your and your filtering as you type is that it's constantly running the search every time you type a character so he's doing the standard. Debounce here, where he's got a timer running, and only comparing and he's saying if between when the timer started and ended, if what I've typed has changed, then run another filter. So that's kind of standard.

A

Well, I think this actually shows a lot of the power of blazer programming model, because I wouldn't want to write debounce code in JavaScript I'm, not actually.

C

It's probably easier engines.

A

Maybe I would be a little worried here. I can keep in my standard, dotnet timer world and and so so anyhow, that's that's kind of cool and he's got a bunch. Other stuff here. Queries across the wire he's doing some things: we're sharing UI components, a lot of cool stuff, so wonderful, stuff and reminder: we've got a chef community stand up as well going on so they're they're, showing off a lot of cool stuff in those another neat blazer thing this is blazer repple. So this is an example here.

A

One is where you can type in your own stuff and do the repple thing. They've also got some built-in ones. So some things like counter demo: they even have here's a forum demo and so this. Actually this is all the code and then, if I hit run, it actually processes and runs it interactively. So that's this one, that's a little bit more complex, but you know this is a full repple thing showing this all so pretty neat stuff one or two more Blazer things here.

A

This is a neat deep dive by Edie into the render tree so explaining how things how the render tree is constructed, and you know showing the abstraction between the DOM and the and the render tree itself and explaining it. So he goes through and explains how it's constructed some nice visuals here showing how it's all fit together and then he kind of towards the end of this talks about, and he shows here how you can construct things and work directly with the render tree through the component base.

A

One thing I have to get upset, though, is here he's got multiple h1 elements he iterates through at one point and creates multiple h1 elements and I don't like that at all, but here he shows building out a render tree and then, finally, at the very end, he talks about a place where you can. If you're directly editing things in the Dom, you can, you can get a conflict, and so he shows using the key or you can set the key to allow for direct tracking of an element so really good stuff.

A

There all right, here's one more lengthy, blazer, post, tons of good stuff here and blazer lately, as so daniel writing about using laser tensorflow and ml dotnet to identify images in this case broccoli. So he goes through. He shows how he creates a tensor flow model and trains it with a bunch of images and and labels showing what those images are.

A

So this is actually just a console application that he runs separately creates a tensor flow model is neat visualizer, which is pretty cool, and then he dumps that into an or a blazer application with ml net and then goes through, and you know show some some additional things. You've got to do, for instance, setting it up with system drawing on Mac or Linux, and then then then there's some more kind of in-depth stuff as far as actually kind of cleaning it up.

A

He does some nice things even as far as like when you upload the image. He does some some nice things where he actually loads in the image and shows a loading indication. So, for instance, here he is loading the image by decoding the binary 60 base64 binary. Here and showing that so pretty cool stuff, it's neat I I, like that ml dotnet works with existing things like tensorflow, and this is cool to see it kind of fully. You know integrated solution there.

A

Well, alright, update from Elan on mobile blazer bindings, not a ton of stuff in this one. Some some things like the update for xamarin forms: four: five simplified more web like syntax. So instead of this button text equals you can have button and and between, and also some things, including in like CSS improvements I, like also that he always calls out a community input. So you know so that's pretty cool too all right coming up very soon.

A

This blazer day, this is on the 18th and there's a ton of stuff there's a lot of great sessions. So, of course, Dan Roth we've got IDI, we've got you know, just all the usual suspects have blazer world. So it's great to see, see this event coming up.

A

Alright. Moving on from the blazer section. Here we have a nice one from Andrew Locke and he's talking about boost filtering with Castro. So he points out the problem that you can, if you're not specifying boost names, you open yourself up for some, so he shows some examples with things like DNS, rebinding, cache, poisoning and password reset hijacking.

A

So some like some things with DNS rebinding. This was something I like, for instance. Here is an example where we have another host name.

A

This is a sub domain and the sub domain is set with a really short T, DL and like 60 seconds, and then it's updated to point to another URL and then they start doing xhr requests so scary things. So what he shows here actually is is a very simple solution which is actually setting your host your aloud host. So by default it's just star, but you can set them directly so and then he shows here by setting that directly he's allowing site a but not site B.

A

So here we got in get an invalid host name so stuff one other kind of secure security related thing.

A

I thought was interesting recently, so Manjula is writing about target underscore blank, and she points out that by default you can set like if, if I have a pop-up and that pop-up can maliciously set window dot, opener dot location and it can redirect to something that looks like so, for instance, if I, if a site has a window dot opener that window dot opener, could then impersonate like a facebook, login page or something and all of a sudden I'm entering my credentials, giving my credentials away.

A

So she points out that there is the no opener rel attribute I, wasn't aware of so there's no open there and then also Fri 11. Your safe are also putting in no referrer. What that does is your underscore. Blank does not allow changing the window, dot, opener, so good stuff, all right. Let us encrypt from Nate McMaster. So a few things to to be aware of here, one is its. Let us encrypt or instead of let's encrypt I, feel.

C

I believe it's pronounced 'la tous à tous.

A

Encrypt that makes sense so yeah. So that's that is an update. It also does the automatic update for automatic request for your certificate at thirty days, which is really nice, so this is this is some things to be aware of there all right, some cool announcements. This was just from today, G RPC web now available. So we previously I think both showed this off on the show and also showed the previous longer blog post, explaining it and so, okay, okay, so so this is so G RPC web offers.

A

Support for, like apps from the browser support from is and app service, which is nice yeah. So what's cool Damien is I, cannot turn out you'll see here, I've got this actually set focus I'm.

C

Sorry now I know: I can do that I'm just gonna. Damn you.

A

I know teams does not follow, does not respect those settings which is kind of annoying so yeah. We.

C

Should make sure you thin feedback yep.

A

That's true um so yeah so anyhow this this is this is release today. So this is very cool, so gr PC web now available. Congratulations, team, hello! Everyone, hey Glenn! What's up also Tim with the web live preview, so web live preview. This is this is an extension or Visual Studio for Windows you can! Oh, my gosh I'm gonna end the community section very early. So with this you can now go in you. Can this.

A

You can both interact directly on.

C

A

Right you can interact directly, but what's also cool about this is that it allows interaction with the our tools, so you can actually go in with browser tools, update things and it'll update directly back into your code in Visual, Studio itself, so very, very exciting stuff. Give it a try and from the feedbacks. Okay. Last thing, very quick one here so up with the asp net core for dotnet 5 preview, 5 update, it's not a huge amount in here very easy to update, and the only update here is reloadable update via configuration for kestrel.

A

So but if you're on the.net 5 train, you should update for the latest previews and I am going to share this link out and I am done. Sharing my screen. So I will go back to this screen and that we're all back.

B

Awesome Big John, you said we don't have any features in a snake.

A

B

Failure I said.

A

The only one that was in shrubs is.

B

Your clean that we don't have features and I've didn't do any work. Is that you? What you're saying I implied.

A

B

A

You called yourself out I just left you alright. So what do you folks have to show us today? What.

B

Do you have Seb I.

D

Have well my crucial orchard core Asti, to which we ship.

D

Yeah, so the idea is that we've worked for too long now on performance and we've improved the system. We are using to benchmark SP, Annette and everything else, to a point that it's reusable by everyone who cares about earth. So we want to to make it more usable, more open, saucy with documentation and just clean stuff. So we made a new repository to to to get it open source and to let everyone use it if they want to make benchmarks on on the web.

D

Apps library web apps, including that net, it's better on that night action. So yeah is today to show you how it works, how we measure stuff and how you can reuse it to measure your own apps or just help. The dotnet team make things faster for everyone, so.

B

The background story is Seb, made sure that that is open-source today and benchmarks, and it was basically over time. It can. I grew organically into this thing that we used to test. I can power on all of our scenarios? I mean track it via the via the power bi. That Damien shows like every now and then on. Stand-Ups are in other places, is how we track our tech employ our progress.

B

So you can run this thing on all of our perfect machines to show perf numbers, as as a team grew as more people got involved, it became really hard for Sepp to run all the process, so he tried to make it easy for everyone to run and that kind of grew and grew more organically and the command line kind of this monster of a bunch of things to pass in so sad bit kind of a reboot um to make it easier to basically run these.

B

These performance tests and I saw it and I said said: we have to make this thing like. We have to make this thing a product because it actually is super useful by itself and we had a bunch of internal teams on other people in general want to test their stuff and we and we were always go and tell them. We have this thing. Does it have any thoughts? But if you can figure it out, you know, then that makes sense.

B

So we made a repo like how long ago, like maybe two weeks ago, on an a4 sub to name it I'll be head of me and we we we named it. We did some stuff, and now we think we have what we think is a super cool product for for perf testing and maybe a little testing too. Let's crash testing, my video frozen still.

E

B

Here is your internet, fine, so yeah so step show what you have first.

D

Of all, what's.

B

The name, what's the name well,.

D

That's ian's forced us to use it so I mean you can I.

C

D

C

Power I said suggestions, suggestions, okay to the preference; and lo and behold, it's the one that y'all ended up using. You can't blame.

B

Me that's cause. We got, we got just annoyed trying to pick names. Nemi is really hard, so.

D

It's called crank I'm, showing my school now. I hope you can see it. We didn't even try.

D

Thank you, and so I just asked my wife to make no sound, no noise, but she said my daughter has a piano lesson in 30 minutes. So I'm very sorry, we'll see how it goes. Yes, so that's the repositories in private. It will be public soon, I keep hitting f5 and it's not panic. It's because I'm David, so we'll see how it goes. So this is the repository dotnet slash crank previously to us. Well, it was it still there, its asp net benchmarks.

D

This is where we are where we have everything, unfrosted sure, but also all the apps. We want to measure and lots of things and too many things that fly or so we made a new repository to just separate what we do in the asp.net team and what everyone should be able to use, which is just the infra, all the tools that we use and there are superior documentation and we are the CIA now has tests and it builds stuff. It's no more like junk.

D

B

Made before because you could break it easily because it had no text, oh good, so we have test no.

D

Yes, so the idea of these tool is to be able to deploy jobs, apps on different servers and let them rappin like a web app load generator a database readies whatever you want on whatever machine you want, and then it will gather some mate that will be returned to your machine to be stored somewhere in a database or in JSON. So you can see the result of the benchmarks.

B

Okay straight jobs right on machines and then the controller will coordinate multiple jobs across multiple machines, gather metrics, the other stats about other jobs, I'm bringing to some in some some form, storage, database or or a file. So.

D

This schema, what we see is that the controller is like to CLI that we talk to agents. Every agent is installed on a machine or just on a single machine. It depends on what I want to do, and every agent can run docker jobs. Does that project or just executable files? Because if you don't want to belittle a job, you might have something that is already compiled, you can send it directly or if you want to run a docker job. You can also do that.

D

So this is gives us a lot of flexibility about what we want to run and to benchmark, and an agent is just the net service that you install on a machine that will be part of the of the benchmarking.

D

So it's a donate web app talks HTTP through the controller, and it will just start your jobs and stop them is the orchestrator and we call it the agent and the controller is a command-line tool that we call crank. This is a crank agent.

D

This is crank and from there we can start commands that will orchestrate all the agents that will start the job and the agent is able to to measure what's happening on each application, like the CPU to memory edge network, swapping memory and things like this and then each app can also return, custom, metrics and that's very powerful feature. Because then you can decide on what information you want your app to to provide to the agent and that will be forwarded to the controller.

D

So the controller will Adhan you about everything in that run, on the different machine, yep.

B

Also, the the agent can do things like collect profiles, so you can imagine running a load test and then capturing an allocation profile or capturing a CPU profile or perfu trace, and it can basically run remotely collect information and send that data back to the controller as well. So you.

C

B

Metrics logs traces, a bunch of different things from from the application, so.

D

When David mentioned that it grew organically, it became a monster because at first we just want to like to run an application and start W. Ok to measure RPS. Ok request pass again: that's all we wanted to do, and then we were like ok, but we need to measure memory to measure. Latency is to measure network usage to measure is the GC enabled on that? Is the seller GC enabled? Or what version are we actually running? Is it like 2.1 or 2.2, or is it whatever and and then we kept adding comments?

D

Yes, arguments to the command line and more and more and more and more features like Indian, we have I know maybe 50 different things that we can measure with these two and then same.

C

Was go on vacation and we'd have to find the wiki page. One.

B

Not all of the magical incantations, the best part was the first version like we had a single benchmark that had all the scenarios and everything's narrow.

C

B

A flag you pass on the command line if.

A

You'd, like run.

B

Kestrel with plain text with this thing, all these arguments and remember.

C

That was crazy. There's.

B

Like whoa, what was born.

C

That the history of this thing is literally I, think I I started building that app and it just had some rudimentary odd casting in the web app itself, and you would say: do you want to run the database test? So do you want to run like the middleware tests and I use some ridiculous, like naive pattern, matching to try and turn middleware and stuff on and off, and then Mike I think built the driver. That would let us send jobs to it remotely when he was working on our performance stuff as well.

C

So then we had like infrastructure. That was our first piece of like always running infrastructure that we could submit jobs to virus sh or whatever. Then, of course, we wanted to get into automation. So then you want things like the build, would automatically push a job to the infrastructure and then give you results back, but then we still want to be able to ad hoc so that a dev can make a code change, build the assembly locally without having to push it anywhere, but invoke a job with their private assembly.

C

So that was something that I know. A lot of people did, or this private copy of dotnet core, for example, so I know we did that for a long time and then we as subsets before we had to start onboarding other teams, because it wasn't just a spin anymore there be it.

C

Someone from the dotnet core runtime team would be working on a fix to improve the performance of one of our scenarios, and they would need to be able to use this tool as well, and so, like you, throw them the one now and they're like what yeah, and so this has really grown into something. And then then we want to start running competitive stuff right.

C

So not only do we run our own dotnet server, but like tech and power we want to be able to track our performance on our hardware versus other frameworks or other things that they implement a different ways so that we can get like for like numbers on the same machine. So it needs to be other. You know, needs be able to run anything. So all that has led to where we are right now, which is.

B

Pretty cool, yeah and so I think one day we're in a we're in a meeting and said had an epiphany that he could rewrite the thing to be super generic and I think he spent like two weeks just scrapping the old, not scrapping, making a brand new agent that we're seeing her. That was a new and improved system and over the last couple of months it kind of got better or better. It's.

A

Interesting Brian Howard from Tech and Harris said conceptually. It sounds similar to the TFB toolkit itself, yeah.

B

D

Yeah, so here I'm also showing the what is the goal of our infrastructure to be able to measure performance in iOS and and shout it and filter it and compare scenarios. So, as I mentioned, we measure, for instance, where can I find some? Where are they? Oh is too much featuring many yes yeah, so we measure different frameworks like no GS rest at clicks go so, for instance, you know you're acting fast HTTP, which is go on the OJS.

D

You need super Spurs and- and we can then check the box season and then see the comparison and the chops.

B

Hunters, so if you scroll down for like done that scenario, so you can see all the counters for all the stats like RPS memory threat, pokey usage, one.

C

Gentoos and what is it normally? That's: a combination of both dotnet runtime counters or Avenger counters or metrics and platform counters, so that we can. We can gather information directly from the machine, that's running it like the underlying OS or we can get it from the process itself or you can deploy another job. That job is to gather metrics from the other job.

C

That's running like you can do all that type of composition and then, ultimately, it all just gets put back into a database, and then you can visualize it whatever you like we're using power bi, but the idea would be if you in and deploy this yourself. The data goes to wherever you want the day to go to go to, and then you could visualize it using whatever tool. You want to write.

A

Under variable that coral I correlate things together, for instance, let's say there's a GC is causing a slow down and per for whatever, like and and see. Okay that was caused by this yeah.

C

So, just by virtue of all these time series being synchronized right, so these graphs here are all showing the results for whatever is currently selected. And so, if you see like a jump, let's have a look on the top left-hand graph. Where it says IPS you can see. We have a step up back in May, so we got faster right and that happened on on May, the 9th May the 8th, and you can see all the metadata in the tooltip, which is a feature of power bi.

C

You can pick other fields to show up in the tooltips for each of these data series, so we can see what Bill debasement at kora was what build of dotnet core. It was what sessionid, what job it was, and so we can look at that and then you can. Then you can hover over another graph like how much memory was being used or what was the latency or what was the CPU or what was the contention rate and you can find the same dot right.

C

The same series start at the same point in time with the same runtime version, and you know the correlate. So you can see right here at around the same time. A fixed obviously went in that made a change to the thread pool because before that series of time the thread pool items was much higher and then it went down, and that seems to have happened around the same time, that the RPS went up right.

B

And this is used, we also track regressions. So if we do have a like a the perfect regressing in some form memory getting higher our CPU getting slower our RPS getting lower, we can late it to commit and dip them and figure out what change will hopefully figure out what changes in a snit core or donate core caused that issue and SEP has a bot. That's these scans through commits that W called the regression, but- and it follows issues when, when the level of the level of various for regression is is has hit some threshold.

B

So here's an example of one, a four percent change right.

B

A

B

A

About this should be hooked up to get blame and automatically like what.

B

B

You'd only have that if you were super confident that this is the problem, but a later time you can give you a diff yeah. It takes some some depth to go figure. Ok, this this change, maybe cause the regression, for the most part is pretty clear, but sometimes it's not obvious and it could be a red herring and.

C

We also obviously have environmental changes from time to time, whether it's kernel patches or your drivers, and things like that. Now, like any good performance environment, you want to try and really coordinate those and make sure you do those in a very deliberate fashion, but sometimes things come through and we don't realize that it was an environmental change that caused it. But we've got a lot better of that. I'd spare to say I think so we haven't really have well. There is one.

D

Well, but I can explain. That's that's fine! That's a good inventory man gunshot, but you can see it here because I'm measuring HF proxy and an HTTP client proxy. So this is a page where we track proxy performance and if I compare H a proxy to a CD clients, you see the two of them. They have an issue and it's not because of the net runtime change that HT proxy went slower. It's actually like a different discussion, but this is how we find that it's an environmental change and not.

C

Be running competitor or non net benchmarks, always as well, because, as I said, it makes it very clear when an environmental change potentially impacted everything and.

D

You see here, for instance, everything you didn't work so yeah there was something going on. Machine was gone or CI. Wasn't her name later.

C

So we're running this on all of our environments. Now right so we have you know: windows. We have physical windows, we have Linux a physical Linux. We have cloud-based Windows and Linux as well, and then we have a couple of other environments that we do ad-hoc right with the like, the from other teams. They have hardware that we sometimes are able to utilize to get some variance or we have a net.

C

We have an AMD machine as well, so we have a a Linux AMD server, so we've started doing comparisons of the epic epic Rome, the new AMD server architecture, and to have a look at how we do on that sort of hardware as well, and we run all of these on and you can see if we've got arm in there as well. So we've got a whole bunch of different physical machines and virtual machines under cloud that we're running all of these benchmarks on all the time so that we can compare these things. So there's.

A

A question from Wraith about tech empower benchmark, running against sequel server. We have no.

C

Comment I've got nothing to to add to that today. Unfortunately,.

A

C

Chase up the appropriate contacts again, but there's a long-standing issue on trying to get that enabled. Okay.

A

Because I was thinking too, especially now that there's the sequel server for Linux or be interesting to compare with that stuff.

D

So I wanted to show some demos how it so this is the end result. But how do we get to these numbers? How do you run a benchmark like I was showing different dashboards because every team in the net might have different requirements like the chief computation team wanted to analyze the differences between achieve completion on enough? How is it good for our peers on a 10th season started grdc, like James, has lots of benchmarks for comparing go native and the net implementation on the server and on the client.

D

So we can combine everything here with different type of payloads or Linux and Windows, and he gets old shots. He wants to draw the perf and of these things and to.

B

Be clear: this is done. It Ananda net code running in the benchmark, so.

D

And I will show you how to run that with the same.

D

E

D

B

That I didn't see it.

D

So here I'm under clone of the crank repository, but you don't need that I would easy, because I have samples there. I want you to reuse. So crank is available as a dead neck tool actually as to that new tools for the controller and for the agent. So if I go on the getting starting by.

B

The way those dogs were built, the diagrams were built with darada I/o and.

A

That's my my Constitution is.

B

To draw super fantasy diagrams to be an architect very.

A

D

So here the net wins told the crane controller and the crank agent, and then you are able to run crank for the controller and crank agent to run the agent. So if I just want to run an agent look because I want to do some tests on my local machine that will run all the application. I can do crank agent and it starts the.net service. It's the net app that will accept any jobs from the controller, and here it's on the local host on five zero or ten pot, and now the the agent is ready.

D

It's waiting for jobs to run. If.

B

You hit the endpoint, what happens.

D

Nothing fancy this.

B

Is the best part of the whole demo? Just so you guys know.

B

That means it's working. That means it's working. It's up.

A

And running, and they tell you in strapped, you need to bootstrap this side up, I.

D

Could use n swag and have alright didn't have a nice cherry.

B

D

Some point: it's open-source, you can provide a peon. Maybe it.

B

Will be open source.

D

So here and now what I want to do is run I will run a web app, an expedite way back. That is on the central supposed to be so. If I look at that I have it there if I open my hello folder in the samples, the hello folder is just a simple meter where standard app that gives end point and returns hello. World okay is just sending returning edward on route path. Okay, and it's also displaying some the net version that is training on, but we'll see later.

D

So that's how it does so who MCS Alex? Yes, yes broach the CEO says it can run on 3 1, &, 5, 0, ok, much targeting and that's rack 5 ornette, 5 yeah.

C

You could change that to the Northshore DFM now, if.

E

You want to take yes.

C

E

Want to break her down, I, don't know man just for you and you will see 3 work, yeah.

E

D

We need a change remember no 2 month ago here. The idea is that from the command line, crank I need to be able to describe the benchmark itself. The benchmark I want to run is made of this application. The web application that I want to deploy somewhere and a load generation tool that we'll use to simulate some HTTP client when some HTTP traffic under web and nationalism. This tool will use the we use, W or K or Bombardier I- think here, I'm using one dot here.

D

So this file is describing everything that we need to do to run a benchmark. I won't use this one actually excuse me: I will use the local one here, which is simple rissalah. So here these file defines jobs, scenarios and profiles. So if I focus on the jobs first, it says I have a job name server and the source is in this local folder, ok and the project I want to run is the cs proj here, and the Redis type text is application sake. So what it says is the application is local in this folder.

D

You need to build a CH file. Remember we can run docker files, CS, proj and executables, and to detect that the application is ready to accept requests. You have to check that the text application start is available, and this is what dotnet apps web 18 I'd, actually output on the console to say that they are ready. So that's how line into you and then here I'm importing another job, which is the Bombardier job, and this job has the same definition as that.

D

That explains how to run my body, which is a console app made in, go to generate some HTTP node. It's like W, ok, but in this example, one body is very interesting for us because it can run on Windows and Linux W. Ok, we only run on Linux. So that's why I'm using my body for these demos, the scenarios is using existing jobs by explaining what you run and want right.

D

So here it says the sanaya name, hello, who will deploy something called an application that will be based on the job server, so the web app I just defined here and then it says, I have a load service to deploy which we base on the job Bombardier, which is defined this file and the variables for this job are that the server port is 5,000 and the path request is slash. Okay, so that's like a docker compose file that will explain what are the components of my system and what to deploy.

D

So this one and then this one another profile will tell the tool where to deploy every job of my scenario, so it will actually define that for the job for the jobs. The application and point will be this machine, my local machine and the load. One will be this machine. So the end point is the URL of the agent that will accept the job that will get the job from the controller and, in my case, I just started imagined on my machine on 50-ton.

D

Okay, the two jobs I will send them to the same local region. So you usually don't do that when you want to do a benchmark, but further the sake of the demonstration I will just use one, my own local machine to run everything and the server URI is a variable that is used by Bombardier.

D

If I open the file will open just to show you that it's the same thing that doesn't job it has variables that will help me define the command line to run given value and the server URI is actually the one that is used to target the application to to send HTTP requests to. So, if I go back there, the server URI is also across and the port will be 5000, which is a default one on this estimate up so now from here.

D

What I can say is crank and then I will pass the configuring file, which is in samples local, this Phi and from there I can say which scenario I want to run, because there might be many in that case is just hello and I will say like a profile, and there is only one which is local.

D

What's happening now, is it's sending the first job top one to my local machine, so here using local folders, so it's using the local folder, it's uploading, the local folder, to the machine and if I look on the agent, the agent got the source code and now accepted the job and is currently seeing that it wants to run in Oetker up 3-1. So it's downloading all the requirements for this tab, which is the runtime 3 1 5, the new SDK, when the latest runtime, the latest SDK, the latest XP net and the latest desktop runtime.

D

So here you don't know in everything and installing everything, so it might take some seconds to to run and how it found. This version is because, if you look at this yes proj, it was accepting 2t FM's that curl up 3-1 and net corrupt 5-0. So by default it says. Ok, there are two of them. I take the first one. This is how it works and then which version so? Is it the one that is currently built on the nightly feed of exponet or just maybe an old one? No, no it!

D

It's taking the latest available publicly available, which we call current and currently.

A

You get up from where do you get those different values and the map says decay to the current, so write a PhD.

D

About that there are feeds and metadata files online on the github repositories and you get and private feeds that will contain these sessions. So in this case it found it was three one file and I will show you how it can detect more versions. So it's what everything it's totally lifting and now it's running you see the application is running and it's starting doing something for bombardier. So Bombardier is also configured to use three one: five. Actually let go up three one.

D

The app that we are building is actually wrapper around my body and both up and our running, and you can see the state here so I can see on the server that everything is running and at some point after few seconds it will stop.

D

I can see some stats going on and I'm not supposed to see that it's supposed to run as a service on a machine and it's stopping the the jobs are and if I go back on my client, my controller I see the results, so I can see what happened job one job 2, which is a server and when the application on the load- and here I see the results from the application and results from the load. What's common between the two is that the CPUs age is provided.

D

I have the memory the build time and the start time very important to make sure how long it to start the application first and then the load itself return more important information, like the time of the first request. How many requests were sent? How many bad response is the min latency, the max latency requests per second and the max of the requests per second, and this is typical from embody. The Barry okay doesn't return you the the max requests per. Second, it's like I, don't know how to do that.

D

Maybe this sample during one second and every second, and they return and mean requests per second and a max requests per second. But when we chart it, we always return the average request when the mean requests per second, which is how many requests by how long. In this case, it was fourteen thousand one 144,000 by fifteen seconds. That's like so that's everything we get from this run, and you see it was pretty simple. I just needed to to run that now.

D

If I was to change the the my local application, it will just send it again: Andrey benchmark it right so from there we can change the scenario to make it faster because run I. It was running 50 seconds of normal and 15 seconds of measurements, so I could do that either in the file by changing some variables, because Bombardier will use a one map generation. So I could say well met like two seconds in duration and.

B

And these names come from the job itself, so the job is self defined variables. So you have to look at the Bombardier definition to understand what variables are used in the in the command-line arguments and now I tell you what what you can set in the section air, it's gonna figure.

D

Okay and it's completely open, because the idea is that one bad year we have an argument or party that says so, if you know Bombardier, it's like a cheap and choda-boy acadia all have the same parameters, so the number of connections and the duration of the run. This is our own. You see one map duration, how many requests to send if there is a rate defined past the rate? So this is, we are templating the command line to run on the server and this way from our job.

D

We can change all these variables like we can overwrite. So there is a default and we can change wherever want from wherever we want, and this wherever we want, is either our own file here or even on the command line. So if I don't want to do that for everyone, because the goal is to be able to share this Nile, but I just want to do it from the command line.

D

I can just do that, and here a crank will look for an existing argument and if it doesn't find it, it will bind it to the name of the job. So if I say load that variables.

D

That duration, it will just change the viable duration of the load job, but it would be easier for me to define a global, viable to say one equal zero. So there are many ways to to define all these variables on the command line on the job as a global variable or yeah, so that that's the flexibility we have we have is that and this.

B

Will work because because Jafra can define I shared things in the benchmarks, armed people and that can it can like can over, have the change files back and for it per run. You.

A

Can just change it in the command line? Yeah, yes, and the job off question, and the chat is how you handle, or do you handle cases where you have authentication required like where you need to pass in a token or handle some sort of authentication scenario? Yes,.

D

So we don't do anything here, it's to your client and server to define how they will communicate. But, for instance, in we've used about you, okay, you can pass a script because of Arrakis opposed to a script, and there are examples in the dubillac a repository on how to intercept every request to do a pre-flight request to get a token and then assign this bearer token on every other request. After that, so the idea would be to have your client support, authentication. Another.

C

Example, you can you create a job and then can you do? We have output variables? I guess is the question I need to ask, so you can repeat that: do we have the support for output variables from jobs? No, because I think if you did have that it'll just do the feature design on the call hi. You could have a job that does a generic pre-flight request job that then emits a token into an output variable, and then you could flow that into the next job.

C

As a header like a more body air support, setting, headers right by.

D

We dist templating, we could just have a helper, too yeah. We could have something to bind the out yeah.

C

Exactly you can actually use a little DSL to do that they would just do it for you. So for.

D

Instance here what what that's, why I wanted to sure? So, if I do application, dot options that display output, because the application itself was outputting, some dashing okay, so here I can configure that to stream the output of my application to my local machine. So imagine you don't have access like I did here to the you can say: okay show me what's hot booty and she live I, see that my application, starting and I, see what it's out cooking. So it fits looking something something.

D

I can see the my local machine, so we already have the so yeah in somehow we could be able to extract some values and then use it for the next job that will not be hard actually so here and now job is run now building. This is for the next job and if I control C now it will stop the client and every agent we just kill the applications that they run. But I won't do that. It should be quick. Now it's done, but you see I D.

D

You can see the output of every endpoint of of your system put it. So that was the first demo. What I wanted to show now is how to run an actual thing on actual machine. So what I have here? I have three different machines: a speaker of clean load and DB. This is an environment we use to benchmark our applications before we look at them on the CI.

D

The Kemper is using an environment of the whole city and we have the same one same machines, but this one is a smaller one with 12 cores on each machine, and this is very useful for us to test regressions without having to to add more true than the machines we use for this.

C

I'll leave the correct name for those machines. Is the asp.net Smurfs lab.

A

D

So to use 3d, so each of these are the major and 20 okay. So instead of bringing my own will call up, agent I will use these ones now and I will use a measure sample, which is the app you were complaining about. The nice part that does everything the beautiful, app and I have a sample like this. So I click trends here and if I open this file its yeah.

D

Now you understand how it works. It's importing the WHI job, the Bombardier job, so I have some choice here we use the very okay because it's doing pipelining and it's very important for us because of the camp where and here in the job I have the a spirit benchmark which is not a local app but something on Gaeta. So we can also say the app I want to benchmark, isn't heated on this branch, and this is the project to be very interesting and here the variables for this app.

D

So this app itself when, when it starts it needs to be defined and time I'll explain that and I was. It was kind of cringe because we are still using all these parameters. What is an eye? What transport, what server, what protocol all these things- and this is the command line to start this up, so this app can do many things to many things. We break it eventually, yes, and we passed a ministry.

D

Then we also have the platform benchmarks, which is faster, but we won't care about it now and the Postgres equal job, which is a book of file, and this docker file is also on repository and we give it a name and the context directory and how we know that Postgres is ready by outputting that and we that we can be fine, let's say the fortune scenario, so the fortune said: I won on tech.

D

Empire is about having database job that we use from the docker image a web application and we'll use the asp.net one and we'll configure it to say you need to run the DB fortunes frozen. Are you using the progress secret arrays? This is the set of arguments that the man was complaining about.

D

We can also define environment variables because we want to pass a connection string and then we'll define a load job using the Bayaka with headers and hitting a specific path on a specific boat and now in the profiles, I still have my local or file, but I have no profiles. I have the one that is using the three machine. I showed you for the application we'll go in the lean machine for the load will go in the load machine and for the database on the one that is cool database.

D

Okay, three different agents, and what special is this machine is that either is that they have a private network, private LAN. So from my network, they are known as these and but internally now or network, with an old networker, an own networks which the the web application is known at as 1000 102 and the database one has 1000 103. So we are sure the the network traffic will be local without any interference from the public network.

D

I just need to verify I'm on the VPN I'm on the VP and otherwise I won't be able to access this machine, and the profile is name s Pinet perf beam, so that from the command line. It's very easy to go into this configuration to select a scenario like fortunes, the profile, the one that will send it to these machines and I won't.

D

Do that same thing: free command and lines free the arguments in the green line and if I start it, which will be the first one one, the first one to be deployed will be database. So, if I look at my agent on the today's server.

D

There is nothing what is a turning birth DB, oh okay, I lost. The connection is completely blocked. Okay, so I know the connection on this. One was bad, but now the load has started also David mention that we see is pointing to what is currently running. So if I open this URL, this is what's the state of the job that is running on this machine, so I can follow even retrospectively.

D

I can see what was what happened on my on a run for each of the agents, so Knights done so it deployed an app on each of these machines ran and then stopped when the last one was done and the last one was the load generation and what we get is the same result, but the difference is the the web application that we running is exporting custom information, so in this case it out put it, but in the console, but as a metric to record the a spirit of the ocean and the runtime version, and in this case it used a 5:01.

D

This is the latest public version of ASP ID 5 preview 305, and this is what it used, and this is the result we get in terms of requests per second 107 thousand requests per second on these machines, slow news, you see using database and it's this machine now, let's say I want to run it on net corrupt 3-1. So what I will do is I will say the application that framework will be met core up 3.1.

D

It will use the current version of net crop, 3 1, which is 3, 1 5 and we'll get the results and we can see if it's faster or not not right.

B

It should be not all right, because.

D

We do a great job when we do the engineers, the ones work on performance I just measure the numbers, because they do a great job. You will see that there is a nice improvement in terms of performance and the fortunes scenario. There should be.

C

It's always dangerous when you call that early said.

A

C

Had to keep people, you have to string them along yeah.

A

Well, well, it's doing this a question about more distributed scenarios and I know in this case, you've already got kind of distributed. You've got like a database, but you know: I was even thinking about cases like project tie and things where you've got kind of more distributed. Things I see a smile. There is that something you've looked at like or distributed scenarios yeah.

D

That's why we remade also this is them so initially before we rebooted this service it can. It could only support three nodes database, which was not actually not for us. It was just a sample that was running all the time.

D

We could not deploy it, but it only supported a server and a client, and that's it so why we made me system is to be able to employ as many as you want on as many machines as you want, so I could deploy ten of them in a single machine, and it's up to you to define to define all the job that you want to deploy so real.

C

Ones that we use, we have, for example, load generator web application database server or we might have load generator web application, back-end API that the web application calls or you might have to load generators, web application, and you know to Redis servers, for example, like there's no reason we have got those type of scenarios that we run already, where we utilize multiple instances of jobs at various tiers which is more distributed. What you're, seeing here, which is just the effectively client-server right so yeah.

D

So Ryan Novak did some studying a few months ago and he was deploying some web api and then he was testing micro services. So you would deploy different apps on different machines to have actual micro service benchmarks to run and something that even David is looking at is to be able to deploy N and, like n, being a big number number of machines. Yeah. It's case, yeah.

B

So we've been doing some work to allocate less memory. That's like a traffic, so I've been trying to do an on Twitter trying to learn how to automate deploying a JVM using terraform for no reason.

B

But the idea is to the install crank agent on on a bunch of a CI containers, that's clients, so I could spin out, like 10 10, a CI ages for client load and then have a big VM. As my server application I could test. The I could try to max out the number of concurrent connections per VM by appending, picking a VM size to doing a test run and then keep doing that over and over upping the size of the VM to get more load into it. Yeah.

C

So it's not just about performance testing. We could also use the to do capacity testing. You know, stress, testing, yeah and there's you they're all the same class they're all basically part of the same family of testing.

E

C

They they typically test for different outcomes. Right and sorry, we used to have a tool in the IAS suite called where the web capacity analysis tool or wk okay, and it was very similar to what you're, seeing here but built on XML and Windows and native code, C++ modules and and really only worked with, is- and this is you know in some ways, a very modern version of that- that's vastly more flexible, yeah and modern because it uses yeah Mille instead of XML, but it can run anything right like it really really can't want anything.

C

So, whether it's you're trying to say hey, given a really big machine as a server. How many concurrent idle WebSocket connections can I connect like? Can I connect half-a-million WebSocket connections or more we'll think about the number of load generating servers? You would need to open that many outgoing connections, one big load generating machine- is probably not going to be either enough or available to you right.

C

But spinning up a hundred container instances on your favorite cloud provider might be something that's available to you, and so that's exactly the type of things that where the the Fowler's gonna look at right now, because when we're doing capacity, testing or load testing, it's a somewhat different scenario to just performance testing, where you say: hey, connect, 256 connections from one machine and then just pump them as fast as you can and measure the outcome right in this case, we want to try and get.

C

We want to get the server to the point just before it fails effectively and see how many connections it was at that point and then we can take. You know a memory profile at that point and then learn what it is. That's taking up the amount of memory for that many connections and make improvements so there's different scenarios, different styles of testing, but this Orchestrator is generic enough that you can use it for both right. What.

A

What I'm thinking here too, is like this is generic enough that obviously this makes sense for you for building a fast frame web framework, but I could see people using this also for understanding overall performance impact of a distributed service at their building like a production, absolutely.

C

And that's what that's, what W cap was originally built for the old one and there's no reason why yeah you couldn't stand this up against your actual application and then have jobs that represent your users or different. You know: load types or user scenarios for your users and then hit the server with a whole bunch of different. You know, quote: unquote, live user code or user sessions and see how it performs right and I.

E

C

Type the thing the Fowler's talking about is imagine. While the job is running, you could issue more like interactively issue more low.

F

C

You can actually observe what happens in real time, so the system looks stable now, with a thousand clients performing a split of user scenarios of tasks. Eight us beat us see what happens if I, if I, add two thousand new clients who come in and try to do task D because, like they hit a marketing banner on some website or something right that type of interactive load testing is the type of stuff that it's often done in the real world for capacity testing of sites.

C

But you need that bursty kind of ability to orchestrate a whole bunch of folks turning up at one point and so I think that's what Farrell has been looking at this last week with trying to automate things like a CIO.

A

So we don't just have to have handsome and tweet it. We can actually, yes,.

C

F

B

So it's also given so before Mac in the olden days, step used to read infrastructure and every benchmark, and this decoupling has less self focus on the actual image itself and individual teams on their benchmarks. Blazer signal are a spirit core just make clients for that custom price.

C

Is that yeah, because it doesn't focus on that's a really good point? This is the the infrastructure layer. It's not the layer that says what type of requests are you going to make and what behavior do you want, like said said, he's got a couple built in he's: got one body.

C

Air and he's got work work, but if you wanted to do more of a user, you know vs has tools for doing like you record a user web session, and then you play it back and scale right against a server those types of user testing or acceptance or stress testing scenarios. This doesn't care about. It doesn't know about those things. It's just a job.

C

You would write a job and you either put that stuff in a container or annex e or in there just does not net project, and then you configure a job to run that right, and so anyone taking crank could then integrate in any other thing that generates the type of traffic or does the type of behavior that they want to, including a sub show before capturing custom. Metrics, like you, can capture custom details from whatever your custom client is, and then that would be collected for you by crank because it has those generic on capabilities.

D

They I wanted to find yes here you see here. This is a GRDC file made by James and he made custom client apps to generate Java C traffic to benchmark the job position. I he's just a console, automating dotnet and then a Yunel file, and then from anywhere. We can simulate that and now the I think this is. The ad team is using this client to test their job PC traffic on any or they copied that some code to make their own client to test your PC traffic to for performance reasons. So.

B

That's and you can imagine an ecosystem of jobs define somewhere on github that you could pull in as like for your own jobs right. So if somebody made a Jerry, C, client or work client Bombardier, you can reference their their thing from github directly in your script, but they'll happen to build your own client for everything. So yeah.

C

Or if you were building a web RTC video application as long as you can write an app that can create a client or emulate client traffic on that you could orchestrate like the glow testing that environment using something like crank. So it really is that layer underneath and you get the flexibility to write the jobs in whatever language you want that actually executes the the various things so yep.

D

So while you were talking I run, two jobs, I render JSON scenario is in the same file that was defined here and I stored them, so to store a job. I used, the output parameter and I saved the first one jason-3 one and the second one for phi 0 I said no I didn't find 50 that didn't okay, so it creates a file like this. If I open it Jason 1 0.

D

It shows me here for each job everything that it computed- and you see here also the SPL conversion, which was written the metadata, which describes every measure it took with description and wanted these all the measurements. Like all the samples we took from every metric bill time, CPU swap everything and- and it does that for all the nodes of your deployment specific to the very arcade from the hosts. First request: silence: okay, so that's Todd, and this is also what will be stored in in sequel.

D

If you decide to not use the bash output but the sequel option, which lets you store the results in signal arrays, that we use for power bi, for instance, or you can use foreign chatting, it will store the result in a JSON document in your database and from that we can also I decide to compare the results. So there is a command in crank, which is called, compare and I can say: Jaden, 3, 1 and Jason 5 0, and it will just show me a table with all the time.

D

Like benchmark net but by pivoting 90 degrees, so because we have lots of metrics and doesn't fit in the screen or isn't 3, so it has to be vertical and and this way so this is a comment to compare already existing measurements. But what you do usually is that you do a baseline, so you run a benchmark and you output it as a something. Let's call it's great baseline.

D

Without your changes, let us how what our engineers do so you do a baseline and you output it with your like, for instance, three, one-off I do and then once you have that you change your app or you change Kestrel and you build a dealer and let's say you change Kestrel. Let's say you.

D

Let's say you are been adapts and you change this trail every day you do that application that output files and then you pass kestrel the journal, like your local file like see color code, Kestrel, Journal, and it will upload this file or in the agent to be benchmarked inside the other tick debugging. So this is super useful for engineers because they can test their local changes.

D

On actual benchmarking servers, but are stable, not they are local at their local machines and from there they can say, compare baseline, the chisel and instead of saving the result which you could see, but we just it will just output this table from the baseline as the column here and the current results in the column there.

D

So they can just do change, being, ran, benchmark change, the run benchmark and and they will see the result of their local changes on the actual application that they are measuring, which can be a split amplication or your own user application to consider that.

D

So that's the stuff loaded engineers used to to track the improvement of the other Kijiji and by the way, if you look at the requests per second between on JSON, which is middleware Russian. So the standard mention between 3 1 and 5 0 and 5 0 3 2 5, not even preview 6 or 7, ok, which is actually faster. You get 22% on this machine requests us again and again sees when when when then the build time window.

D

B

You show the collect counters feature so.

D

Many things to show so here so many years so here the idea is that I'm running it on Linux, but if I want to switch on Windows I can just say: win here, use a different profile: okay, because I have defined a profile that is using a Windows machine.

D

I could compare women Windows to Linux if I want II and I use the latest here, and that corrupt phi0 actually got the letters, but they go 5-0 and you can say, collect, select me that application that I am sorry I, don't have it I need to look at the you.

A

D

Know memorized I.

A

D

Know but there is an argument: I don't know.

B

D

B

D

What you know why we do something guys first because I'm a dad so controller with me.

D

Here I want to do collect. True I want to do that. King.

A

I'm just disappointed you didn't have that all memorized yeah.

D

We do that so what I'm saying is that I want the application? Well, the agent that is running the application to collect a trace- and this is a collect arguments I want to send to the the application- is creating a trace. This argument is from paraview because I'm targeting a Windows machine, so these correct arguments would be for purview and if I was targeting a Linux machine, I would I could pass the ref collect arguments, earth /f, correct arguments, so this is the native traces that we get miss collect true and you receive as happening.

D

Then we have another option which is done net trace, so I can say that options that been a trace and then, if we use the new dot that trace tool to take the trace and we'll have the manage trace and what happen is that the server to the trace using purview and the driver is now downloading the trace? That's why I have an ETL file here? Ok, who knows how to stop our view and to collect and to stop on it? No here it just just say: I want the trace.

D

So what I just need to do now is a take. Well, I can just cook it at path, start per view. A.

D

Folder and open that and.

D

You should see the JSON sterilizer and the thread tool, so if I select now what is the app benchmarks, which is a PI just run, and we can see it just really slow and there is a meter where- and this is a just- a meter where okay, which is invoked so we can then go there and see what's happening inside so you can see. This is just you just got a trace that you can then analyze to that's for you and that's how you pass custom arguments and what David.

A

Was asking what's the one is set similar on on Linux, 2d, trace, ores or something I get.

D

A

Is different yeah so.

D

Yeah I'm next, you were just perfect, perfect, perfect, perfect.

B

There's perf there's like DTrace, that's more of a freebsd thing, but I think it got for it to the clinics and there's uh there's one more called something of it care of the name of it. This one call I, think beeper yeah something else so.

D

What David wanted me to show you is that thing, so what we do also is that we can collect donate counters, which is a thing that was added 3 into 2 and.

C

Yeah 3 for real it was kind of broken into 2.

B

Do you have to pass collect counters true and the providers.

D

So if you just do that, it will grab all the system run time counters the default system that run time controls- and this is what what we saw in the chest with the GC allocations. Gc hips thread, pull stats, looking contentions and so until super you sweets like looks like the framework and the different libraries can provide whatever and provide whatever numbers they want to expose to the app or to the agent to track them and every system every subsystem can be find their own, and this is actually how we expose also our matrix injections.

D

So here you see, I have much more information so on the application itself. I know I know have all these counters that provided some data. It also provides a CPUs age, which much is almost what could be tracked. Memory usage this in GCC general, but just information about allocations, tip size, exceptions per second. All these things are super interesting and because each system can provide their coil contours here, if I passed, we just used the cast reward up to new I need the new ration 35, which one is the newest I, think.

D

Yet it's raising us so if I do that, it will actually use Oh what.

E

Did I do it's kind of providers.

B

Or provider she's.

D

Very interesting because it says that I can't I, don't know this property, so here.

B

How's, it guess, by the way.

B

So profiles counters la features, Tracy's.

D

Save Wizards speak Russian, never.

B

Has been in his cave Italy working on this thing for a book, how long hello? No maybe six months, the new version.

D

Yeah we started around September I, think yeah.

C

A

Poor folks out.

C

Of their misery, everyone on the.

A

C

D

On so you see here we get the HD Edco counter, so requests per second, which should match approximately what nobody sees. Ok member of request. So what's interesting, is that a spinet counted, 1.6 million requests and the benchmark said: I just sent 1.1 million requests, so there isn't much yeah, that's interesting, because when I did that with specific number of requests on Bombardier I think this is this is about yaki. But if I use my body it matches exactly the number of requests, so that's suggest so.

B

Work is buggy is first thing, because that computer.

D

B

D

Something are so interesting is that you see here I'm using justify, which is the latest public session. Let's say: I want to use the latest nightly build of a spinet Corin runtime I can just a application and I don't change the framework that I can change the channel, and here I can say ang and get the result using the latest. The latest latest available bits on the fields which might have been just built and a word ago.

D

Maybe this is how we change direction and I can you can do the same thing under each framework, so super useful, and not only can you define the channel to just get the latest version or the current a big version, but you can also set a specific version. This is very important when you benchmark, because let's say you do a change in Castro, you don't want the runtime to change every time. You run a benchmark. You want to use a stable, well yeah, stable, runtime, a meaning one that doesn't move.

D

So you will run the benchmark and copy-paste the version of the runtime and then, instead of using edge for the runtime, you will use this dashing and you can then change the experiment. Forget that you want to to measure okay. The goal is that when you do two runs, you might not run two different versions of the runtime, because there was a new build available and, and now you just measuring it, on the wax.

A

D

Thank you, and you see here night season, preview, seven and a built from a 16 which is today v build under 16. So today this is a fifth build that were using from the runtime and same thing from a speed net. Okay, so super useful to to track the performance of estimate, maybe also on your own apps. Maybe you want to see what's the impact of the current SPI dashing on my app I know for us it's pretty soon, because that's how we do the things we want to track.

D

The latest version of everything but depends we.

B

Had a question about benchmark on that I know: it is something special there, but it may not be in the new version yet so.

D

The old version, the old drivers- support when the server supports that the agent supports it, but the driver, the new one, doesn't know how to download the results right now to do that. So I haven't posted that, but to show you because this graph here is using the old driver, skill and the EF core benchmarks, they are micro benchmarks, they are benchmark, done and match box and each result here is a class. So, yes, it's supporting.

D

We can run using exactly all s increments by changing the frameworks by yeah, doing gonna planters, but doesn't make any sense in this case. But, yes, you can run benchmarks continuously using benchmark net, and this is the kind of results that you can. You can get you get the same results it's just that. The way they wanted to show me was to show a chart or a class.

D

You could also have a tree view with all the classes and all the methods and just shake them as much as if supported not in the new, so crank it serve, doesn't support it, but crank agent supports it. So it's just. We have to bought every feature to the new. The new controller, something I didn't mention are super useful for micro-services is that you can say application and then restrict CPU, CPU sets and memory on the server.

D

So let's say I just want to run with one call: I can say CPU set or to take the core number, 0 or 0-2 to say cost 0, 1 and 2, and if I run that.

B

B

D

You it's not here, guess.

C

Not so the answer was actually no. Oh.

D

You learn well.

A

We're about at time and I realized. You have a thousand different things to show off here. This is cool, but um so some kind of wrap-up questions there's one so I had mentioned like the micro service and ty as an option. There's a specific question: can you use this directly with ty or is it more like well ty, you could configure your environment with Ty and then you'd wire. This up. Neither.

B

They're, both today very similar in how they launched launch I'm, mind a process, but there's no overlap today. Ok.

A

I guess one other main thing: people are just asking like you know when and how can they start using this as well soon, hopefully, if.

B

We get lucky this week. My boss has to like hit a proof and then it'll be it should be ezel and.

A

They mentioned to that your that there's a docker container available to you right or there will be that true.

D

There will be, there is a look of file that you can use. That's how we deploy this thing here. We do the curl file, but yeah. We want to build a docker image, so you just have to reference this image. I'll run this image. You don't have to build anything, but there is a look a file at least.

A

Okay, anything else enclosing where which people you show the the expected repo I guess, and then you know we cou. How else will people find out about when this goes out? I guess over? We.

C

Will the wazoo all.

F

C

Will know station.

F

C

B

C

B

We sell seven good luck was.

B

A

Anything else in closing.

C

No way way over time, yeah.

A

Yeah, ok, this is super wonderful. This is really cool, so I think you know I love the the I love the side of it. That I know that my applications keep getting faster magically because I write them in because I picked a winning web framework, so that's cool, but then it also is neat to think about like opportunity to to benchmark my applications to you know I'm, so others very good stuff, a job said well, alright, let me see, can you unshare your screen and we can all wave goodbye? Oh.

A

Shoot- and this is again the thing where I don't have I, we so Damien figured out how we can do dramatic, zoom out and I have not configured it. Yet we will get backlog.

C

Maybe we'll get it for the upcoming Hanselman and Damien show we'll.

A

D

C

D

Down yeah, it works in teams. If I show my screen, I control them all.

A

Wonderful, alright thanks. Everybody.