GitLab Feature Walkthroughs, 23 Mar 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Loadtest Project Demo

Description

In this video Cristiano Casella, a Technical Account Manager at GitLab is walking us through the process of setting up a load test in a GitLab project using LocustIO.

A

A

Open the monitoring in another page.

A

Okay, I open this staging.

A

Okay, so this is the application that I'm going to check to test.

A

From the environment, variable is the final here.

A

Sorry, the the URL I get it from the outer loops here. I define the client and the duration of the test and I get the environment variable URL environment from the from the pipeline.

B

So for for us to use this, we need to take this project when we have an app.

B

Yes, we need to take at this stage for no testing and we need to take this.

A

Exit RIA, as you can see here, I'm going to use the environment URL to to fill this variable.

B

Okay, okay and.

A

We can obviously define a default value for low test, client and duration.

B

A

Yeah yeah, you define a default and if the user were to customize have the variable, if not, we have other four answers.

A

Which is the you're using them? Second,.

B

A

This is the old pipeline I'm just going to give a look. The URL.

A

Okay, the staging is still not existing waiting for the deployment.

B

Can you look at the look at the project files yeah.

A

B

A

B

Python is the one we should take the entire project right and you added just the pipe.

A

This file is including the configuration for the test in this case I'm just getting the the index page page.

A

Basically, this is getting this file and you are able to define a specific set affection, for instance, in this instance, I make a login using a username and password I asked the index and they about page. So you can define diverse page and every section that you want to do. For instance, you can make a post or everything else, and this is what you're getting in the end you're, getting all the test: the number of slaves, the users that you are trying to replicate and the RPS are the reply per second.

A

That is basically your score. More, you get an ID value, a more your platform is responsible and you can trace, obviously, the failure and everything and you can export it data. So, for instance, if I remember well, it's possible to expertise as Jason, and we could include this, for instance, in our merge request or in a dedicated graph or dashboard.

A

Okay, we have our builds.

A

And we shall have now our station. Yes, it is, and this is my staging application- is creating the certificates.

A

Okay, so my fishing application now it's working and from the matrix I will start in few minutes to see the matrix from this environment to.

A

Now, looking at a pipeline, I have the load test so and also it is the old one.

A

Okay. So the look to the log. The load test is loaded. He is carrying the URL from the environment, and here you can see all the requests with the score that we are carrying. So it is not so good for that, because it's a verbose enough fur to understand what is happening to your environment.

B

A

This will continue for or two minutes actually because I set it up a hundred twenty seconds and if I connect to the matrix, as you can see, I can start to trace the usage. We still have to wait for a while.

B

Can you go for the locks ii just.

B

Know in the lobes of the go back.

A

B

So, underneath the operation type on your left side, sorry.

A

B

I just want to say something: yeah.

A

As you can see, you can see the HTTP requests made that from the la host site. This is the first request for the certificate and that you have just the HTTP plan request.

A

That's me and the rest is all the the locals quiet.

A

Okay, as you can see, there is the load that now is stable because I'm keeping the same amount of of request.

B

A

B

Minutes after two minutes, you see.

A

Okay, now, for instance, it got an error max retry exceeded, so there's been some failure. The point could be that at the beginning, the inquest certificate was under creation, so we have to deal with the first creation of certificate for the environment or to keep the staging in plane. This is something that we have to define if I going to run a game that the test we we have less error. But the point is that we collected anyway the video event metrics.

A

So, for instance, regarding this period, I know that I received this amount of requests with the average in the maximum time. So I know that, for instance, my application is able to reach 12 requests by second. In this case, what I can do is to go to my application in the staging environment.

A

Actually, my application is just one note: I can, for instance, scale.

B

A

Three notes and do the same test again.

B

In the future, yeah.

A

And so, and now I don't have anymore the problem of guaranteeing a certificate because they still see there is not renew it and I can see the difference to understand how much the vertical or the horizontal scaling, for instance, is impacting. My I could, in with the results I matured in ochre I, who deal with the number of thoughts or everything related to the logic inside application I, for instance, in a complex environment, I could try to scale differently, front end or backhand, or the data base in giant.

B

A

You can see sorry very.

B

A

Thanks, as you can see, the client start, the locust client started to make the request and in the request, but second, the column. You can see that trying to go fast and faster and we will reach at one point the the maximum target. The only limitation in this case it's that to have a real test.

A

I, should scale out also the lowest, and this is something that we need to work on it because from the pipeline, using just as a dogged job is not easy because usually the daughter master and the doc enslave are two different kind of darker image, which will different kind of configuration.

A

The host is providing an uncharged, so, for instance, we could manage this by half, but we need to understand how to integrate this. If we want to scale up the testing in.

B

Giant like multiple users hitting your endpoint.

A

B

User single, using you, yeah I, think.

B

It's enough, we just want to create a load, we're not communicating it in testing environment. We just want to see how we can yeah stressed environment, make sure that this matrix will yeah.

A

For my experience, my use case cloud provider, it wasn't interesting as you create the full amount of the user base. Also because was impossible. It was more important to understand. I use the introduction, sixty four hundred thoughts on ten pots. How many Lord I can push there and I make you know the ratio between the two different environment or adjust now I have this performance if I change this this piece, if my application, this is impacting my performance.

A

So for my for my experience is not so important to have the large number of slave or slow host, we can also deliver an MVC with just one note has I made this in this case.

A

It's totally different from the from the from the pulpo's of the test. So now the job is succeeded. I didn't get any error as expected, because the the endpoint was already there and if I, look at the environment, matrix I.

A

Can see the difference between the two different situation, so in this case, for instance, I can see that the number of tougher memory obviously is increased, but the memory by pod is decrease it and if I go to look at, maybe I can give more.

A

Okay, if I look at the ingress, for instance, I- can see that here. This is really interesting. I know where I make the change, and this is really important in this case, so I can see that I have some difference, but is not changing so much the the answer. So in this case, I can understand that I don't have the pod in sufferance smashes. So with just one note.

A

So, probably if I want to stress more, the pod in this case, I need more slave clients, but it's something that we have to test case by case. There is not just one answer also because actually I asked it to get a blank page, so there is no load inside the pod. It's just a really bad page, but if you think to have club complex application, obviously the test they have to be made or something that is involving the the computing part.

B

What, if I, want to break that up, if I want to create a can I create amount of load that eventually will yeah.

A

But I need slave with, with just one pod I think it is not yeah.

A

What what we could do to have a simple situation instead of having one center master without the slave, is to just replicate the single node. I am NOT going to have a eunuch dashboard inside la justo, but from the metric side that I'm collecting I'm going to have everything. So, for instance, what I could try is.

A

Yeah we we need to define the number of the replicas for the lowest dollar image.

A

So, as mentioned, I think that, obviously, to deliver this, we need to spend some times adjusting something but I think that it's an integration that we could achieve with not so much effort.

B

Now we can, we can. We can first walk on yep, you just show us and if we will see the tool Nate, probably, as you mentioned, to add more replicas to local, so we be able to create more stress to down break the app we'll do it. Maybe maybe we don't need to because probably have a different way to doing what we want to do, and this can help us just spike on the metric make sure the metric system yeah.

A

I thought another approach that we could use, but for the future, I really trust MVC, especially in this kind of case, is to deploy locust with the numbers like that. We need from the github managed app using how and we can try to use the locust API from the diff diverse job to commit. You know the job to the to the locust installation.

A

This could be another approach, so.

B

You create more instances now.

A

Just that we can try to create a central distance from the github management apps and where your pipeline is running is just triggering the mini instance.

B

A

This case, probably we are able to manage better the number of slave and everything related.

B

Well, thank you. I have outbreak in like four minutes.

B

But thanks for showing us this demo, we will take into consideration what we're trying to build our environment. Try to create basically the environment, so your skin will simulate some sort of problem, so the user can go to multiple steps and try to find what's the problem, so it involves multiple parts. One of them is spiking in the matrix and is something we can try. So thanks yeah, thank you. I will share the recording with the monitor team, maybe in case anyone finds it. Yes, Paul you're.

A

Welcome, thanks to you, I copy the maintainer access to the poet, so you can do everything and just let me know in case okay thank.

B

A

Have a good day.