There are lot of steps that are involved in creating an application in Kubernetes. First, you need to develop your container, get it through QA, and then deploy it. In some organizations there are even more steps like security scans and peer reviews that need to be added. One of the steps that many companies decide to leave out until very late in their deployment process, if they even include it at all, is kubernetes load testing.
Kubernetes load testing is an important part of developing an application. You get to validate that your application can withstand the load that you are expecting, minimizing the risk of your application suddenly failing because you experience a spike in traffic. Like if you’re a webshop on Black Friday, or you’re a SaaS company that suddenly gains a huge customer. Increasing your revenue only for your systems to fail is a scenario nobody wants to be in.
With proper load testing tools this can be avoided, and in this article you’ll be introduced to two popular choices: Speedscale and Gatling. These two tools will be compared in five different categories: Ease of setup, development experience, integration into CI/CD, documentation, and pricing.
Ease of Setup
Speedscale provides you with a quick start guide when you log into the web dashboard. This quick start guide stays at the top of the screen until you finish it, but no need to worry, it’s quickly finished as it only has three quick points. Should you need it again at any point down the road, it can always be easily found in the left-hand menu.
The first time you log into the dashboard it’s very helpful, as it shows you the two ways Speedscale can be installed. You can either use Helm to install the Kubernetes Operator, with the quick start guide showing you exactly what values need to be set. Should you decide that Helm isn’t for you, you can also install Speedscale’s CLI tool speedctl, which is the other option provided by the quick start guide. With speedctl installed, it’s as simple as executing speedctl install and from there, speedctl will take care of the rest.
Once you’ve followed either of these options you will have Speedscale installed in your Kubernetes cluster. From here, you can instrument your services with the Speedscale sidecar and start running load tests.
Getting started with Gatling starts with figuring out what kind of Gatling installation you need. There are three options to choose from; Cloud, self-hosted, or "Marketplaces". Marketplaces means either Azure or AWS. Getting started with Cloud or AWS/Azure is fairly easy as it’s just a point-and-click installation through the respective platform’s interface.
Should you want to self-host then you need to first set up a system with the proper JDK version installed, then you need to set up Cassandra to store your data. As part of this, you need to make sure your network is configured properly. With Cassandra set up, you can install Gatling.
Note that the previous paragraphs detail what it takes to install Gatling Enterprise. You can also install Gatling locally, in which case you will only have to worry about installing the right JDK version. However, it’s unlikely that you’ll be using a local version in an enterprise environment.
Comparing the two, it’s clear that Speedscale has the easiest setup. There are two different installation options, both of which are scripts that take care of everything for you. With Gatling it’s a much more manual process, with the installation in AWS/Azure even requiring manual work. Speedscale requires you to have some prior knowledge of Kubernetes, whereas Gatling requires you to know about Java and JDK installation.
Developer Experience
Speedscale can be a bit confusing when you first get started with it, as there are some specific terms you will have to learn, like what a test config and a report is. However, that won’t take long, and following the quick start guide you’ll quickly have your first test running. Because Speedscale works by adding annotations to your Kubernetes Deployments, anyone with prior Kubernetes knowledge should be able to get started working with the tool quickly.
As soon as you’ve instrumented your services with the Speedscale sidecar, Speedscale will start recording and storing traffic. This traffic is used as the base of your tests, so at any point you can easily create a test based on actual traffic that’s been processed in your cluster. More on that in a later section.
A test in Speedscale is easily created by first defining a snapshot, which represents a period of traffic. From here, you need to either patch your existing Deployment in Kubernetes, or deploy a new Deployment. The key thing is that either of these two needs to have some specific annotations, like so:
annotations:
replay.speedscale.com/snapshot-id: "1e4c2995-9acb-4fa7-af1b-9c757e17ed55"
replay.speedscale.com/testconfig-id: "standard"
replay.speedscale.com/cleanup: "inventory"These three annotations define what snapshot it should use, i.e. what traffic needs to be replayed. The test config defines how the traffic is replicated, with "standard" being a one-to-one replication. There are also test configs available which will replicate the traffic 100 times, or you can create your own test configs. The cleanup field will instruct Speedscale to return your cluster to the original state it was before running any tests.
Once you’ve gotten Gatling installed, you can use a recorder.sh script that is installed alongside Gatling. Running this script will open a GUI where you can configure options like the name of the test, a package name, whether the load test should follow redirects, as well as a few other options. Once you click "Start!" in the GUI, it will act as a proxy and you can now visit your site in your browser. Remember that because it acts as a proxy, you will first need to change the proxy settings on your local system to be "localhost". Now, start clicking around on your website and act like a user, the recorder will record all your actions. Once you click "Stop" in the recorder interface, the "Simulation" will be recorded in the folder user-files/simulations/<package-name> of your Gatling installation. If you’ve named your load test "FirstLoadTest" it will be saved with the name FirstLoadTest.java, if you’ve decided to use Java as the basis for your tests.
From here, you can take a look at the test file to see exactly what the recorder has recorded. To run your test, run the gatling.sh script that was installed alongside Gatling. Doing so, Gatling will check through the user-files/simulations folder and list out all your simulations, which it will present in a menu like this:
Choose a simulation number:
[0] FirstLoadTestIt’s tough to say which tool is the easiest to use, as both tools are fairly straightforward once you know how they work. However, it’s clear that Speedscale is by far the most modern of the two options. It provides you with a proper CLI to use, rather than different scripts that need to be run. Should you want to use a graphical interface, Speedscale provides you with one at app.speedscale.com, where Gatling has to be hosted somewhere.
Speedscale also makes it easier to create realistic tests, as it collects data directly from your services. With Gatling, you have to either use a recorder script and generate the data yourself, or write a Java file where you define your test.
Integrating into CI/CD
Speedscale is easy to integrate into any CI/CD pipeline seamlessly. When you’ve run a test a report is generated, which has a status field. This status field is an ENUM, defining one of a list of options. In a CI/CD scenario, this means you can deploy your test, wait for it to be done, check the status report, and from there make a gated check. One key thing to know about Speedscale is that you can define assertions as well as define a load test.
For example, you may decide that it’s not enough to know that your application didn’t crash, you also want to make sure that all requests return a status code of 200. This is possible with Speedscale, and is something you can include in your CI/CD pipeline. As of writing, Speedscale has official documentation on implementing CI/CD in a few different systems, however all options are showing you how to implement a script in a specific system like GitLab CI/CD. Because it’s just a script, you should be able to implement Speedscale in whatever CI/CD system you are using.
To implement Gatling into a CI/CD provider, you first need to make sure you have all the dependencies like Java installed. The second thing to do is to figure out a way to deploy the software that’s in your Pull Request. This can easily be done via Helm or some other software like it, but with Gatling it’s something you will have to manage yourself.
If you are using the free version of Gatling, you will have to rely on third-party resources and workarounds to use Gatling in a CI/CD pipeline, as there’s no official documentation. If you are using the enterprise version, you can find documentation on using plugins in Jenkins, Teamcity, Bamboo, and a short post on how to do it manually.
There’s no way around it, Speedscale is the clear winner when it comes to integrating into a CI/CD scenario. It’s easy to get set up, it’s easy to check a report for whether the load test has completed successfully, something that’s not clear how to do with Gatling, and there aren’t multiple versions you need to keep track of.
Documentation
When you look at the Speedscale documentation you won’t be blown back at a polished product, like what you might see from huge corporations like Microsoft or Google. However, you will find what you need. Clicking through the sections on the left-hand menu you will find precise and succinct information on what you need to know.
Everything is structured in a way that’s understandable, and if you can’t find what you’re searching for by clicking through the menus, you will find a search button in the top right corner which can help you get to what you need.
On the surface, Gatling has a documentation page that looks like what you’d expect from a corporation. It has many sections, long pages, and tons of links to other resources. However, taking a closer look and reading through something like the installation instructions, it becomes apparent that not much focus has been put into the documentation. You’ll often find that the wording is a bit weird, and at times the way everything is structured can be quite confusing.
Also, as mentioned previously Gatling has three different options; local, self-hosted, or cloud. These three options have three separate tabs for documentation, so you need to be aware of what type of product you’re using.
Pricing
The basic version of Speedscale is free, and is essentially a trial of the product. With the free version, you get access to the CLI, TLS decryption, and visibility into all the protocols supported by Speedscale. This is mostly used by those who want API observability, API debugging, and contract testing.
Moving up, you can get the pro version for $500/mo on an annual contract. With this you get all the functionality described in this post, like a WebUI, load generation, service mocking, reporting, CI integration, as well as other features like chaos testing. This is intended for those wanting to do API load testing, automated mocking, and complete CI and GitOps integration. Should you need anything beyond this, like more storage or additional security controls, you can reach out for an enterprise version at a custom price. Both the pro and the enterprise version offers a free 14-day trial.
Gatling is free if you download the basic local version, otherwise you will have to pay a license. Should you want to self-host, the starter license is €400/mo billed annually. However, the officially recommended version is the "Silver" license, which is €625/mo billed annually. The most expensive self-hosted license is €2500/mo.
Should you instead want to use the official Gatling Cloud option, it starts at €69/mo, however the officially recommended package is €276/mo, with the "corporate" option starting at €1380/mo, if billed annually. Billed monthly the lowest package is €99/mo and the recommended option is €396/mo. The corporate option still starts at €1380/mo.
Lastly, there’s the option of installing from the marketplace on AWS or Azure. The basic package starts at $3/hour, the next one being $9/hour, with the "high-scale" version being $19/hour. It’s unclear how many hours Gatling considers to be in a month, but assuming it’s 730 like AWS and Azure do by default, the cheapest option will then be $2190/mo. This is excluding the underlying hosting cost of VMs in AWS/Azure.
From this, it can be determined that there are options in Gatling that are cheaper than Speedscale, however Gatling is trying heavily to push the options that are more expensive. Also, Speedscale only has three different options to choose from, where it can be a much tougher task to figure out what Gatling package you will need.
Conclusion
All in all, it’s clear that Gatling has its place in the world of load testing, however it appears it’s not in the world of modern engineering. Gatling seems to be much more focused on the big enterprise companies that haven’t optimized the development workflow in the past decade.
If you are looking for a tool that can work with more modern technologies and principles, it’s more likely that Speedscale is the choice for you. On top of being able to integrate directly into Kubernetes, it’s also easier to integrate into your workflow like CI/CD pipelines.
If you’ve yet to decide what load testing tool is the right one for you and your Kubernetes infrastructure, check out our other comparisons.
