Skip to main content


From streaming to studio: The evolution of Node.js at Netflix

By Blog, Case Study, Node.js, Project Update

As platforms grow, so do their needs. However, the core infrastructure is often not designed to handle these new challenges as it was optimized for a relatively simple task. Netflix, a member of the OpenJS Foundation, had to overcome this challenge as it evolved from a massive web streaming service to a content production platform. Guilherme Hermeto, Senior Platform Engineer at Netflix, spearheaded efforts to restructure the Netflix Node.js infrastructure to handle new functions while preserving the stability of the application. In his talk below, he walks through his work and provides resources and tips for developers encountering similar problems.

Check out the full presentation 

Netflix initially used Node.js to enable high volume web streaming to over 182 million subscribers. Their three goals with this early infrastructure was to provide observability (metrics), debuggability (diagnostic tools) and availability (service registration). The result was the NodeQuark infrastructure. An application gateway authenticates and routes requests to the NodeQuark service, which then communicates with APIs and formats responses that are sent back to the client. With NodeQuark, Netflix also created a managed experience — teams could create custom API experiences for specific devices. This allows the Netflix app to run seamlessly on different devices. 

Beyond streaming

However, Netflix wanted to move beyond web streaming and into content production. This posed several challenges to the NodeQuark infrastructure and the development team. Web streaming requires relatively few applications, but serves a huge user base. On the other hand, a content production platform houses a large number of applications that serve a limited userbase. Furthermore, a content production app must have multiple levels of security for employees, partners and users. An additional issue is that development for content production is ideally fast paced while platform releases are slow, iterative processes intended to ensure application stability. Grouping these two processes together seems difficult, but the alternative is to spend unnecessary time and effort building a completely separate infrastructure. 

Hermeto decided that in order to solve Netflix’s problems, he would need to use self-contained modules. In other words, plugins! By transitioning to plugins, the Netflix team was able to separate the infrastructure’s functions while still retaining the ability to reuse code shared between web streaming and content production. Hermeto then took plugin architecture to the next step by creating application profiles. The application profile is simply a list of plugins required by an application. The profile reads in these specific plugins and then exports a loaded array. Therefore, the risk of a plugin built for content production breaking the streaming application was reduced. Additionally, by sectioning code out into smaller pieces, the Netflix team was able to remove moving parts from the core system, improving stability. 

Looking ahead

In the future, Hermeto wants to allow teams to create specific application profiles that they can give to customers. Additionally, Netflix may be able to switch from application versions to application profiles as the code breaks into smaller and smaller pieces. 

To finish his talk, Hermeto gave his personal recommendations for open source projects that are useful for observability and debuggability. Essentially, a starting point for building out your own production-level application!  

Personal recommendations for open source projects 

Metrics and alerts: 

Centralized Logging 

Distributed tracing 


Exception Management 

Node.js Package Maintenance: Bridging the gap between maintainers and consumers

By Blog, Node.js, Project Update

This blog was written by Michael Dawson with input from the Node.js package Maintenance Working Group. It was originally posted on the Node.js blog. Node.js is an OpenJS Foundation Impact Project.

Image for post

A while back I talked about the formation of the Node.js package maintenance Working Group and some of the initial steps that we had in mind in terms of helping to move the ecosystem forward. You can read up on that here if you’d like:

This blog is a call to action for package maintainers in order to help in one of our initiatives and to move it forward, we need your help.

It’s been almost 2 years and we’ve been working on a number of initiatives which you can learn more about through the issues in the package-maintenance repo. Things never move quite as fast as we’d like, but we are making progress on a number of different areas.

One area that we identified was:

Building and documenting guidance, tools and processes that businesses can use to identify packages on which they depend, and then to use this information to be able to build a business case that supports their organization and developers helping to maintain those packages.

We started by looking at how to close the gap between maintainers and consumers in terms of expectations. Mismatched expectations can often be a source of friction and by providing a way to communicate the level of support behind a package we believe we can:

  • help maintainers communicate the level of support they can/want to provide, which versions of Node.js they plan to support going forward, and the current level backing in place to help keep development of the package moving forward.
  • reduce potential friction due to mismatched expectations between module maintainers and consumers
  • help consumers better understand the packages they depend on so that they can improve their planning and manage risk.

In terms of managing risk we hope that by helping consumers better understand the key packages they depend on, it will encourage them to support these packages in one or more ways:

  • encouraging their employees to help with the ongoing maintenance
  • provide funding to the existing maintainers
  • support the Foundation the packages are part of (for example the OpenJS Foundation)

After discussion at one of the Node.js Collaborator Summits where there was good support for the concept, the team has worked to define some additional metadata in the package.json in order to allow maintainers to communicate this information.

The detailed specification for this data can be found in:

The TL/DR version is that it allows the maintainer to communicate:

  • target: the platform versions that the package maintainer aims to support. This is different from then existing engines field in that expresses a higher level intent like current LTS version for which the specific versions can change over time.
  • response: how quickly the maintainer chooses to, or is able to, respond to issues and contacts for that level of support
  • backing: how the project is supported, and how consumers can help support the project.

We completed the specification a while ago, but before asking maintainers to start adding the support information we wanted to provide some tooling to help validate that the information added was complete and valid. We’ve just finished the version of that tool which is called support.

The tool currently offers two commands which can be used which are:

  • show
  • validate

The show commands displays a simple tree of the packages for an application and the raw support information for those packages. Much more sophisticated commands will make sense to help consumers review/understand the support info but at this point it’s more important to start to have the information filled in, as that is needed before more sophisticated analysis makes sense.

The validate command helps maintainers validate that they’ve added/defined the support information correctly. If there are errors or omissions it will let the maintainer know so that as support information is added it is high quality and complete.

Our call to action is for package maintainers to:

  • Review the support specification and give us your feedback if you have suggestions/comments.
  • Add support info to your package
  • Use the support tool in order to validate the support info and give us feedback on the tool
  • Let us know when you’ve added support info so that we can keep track of how well we are doing in terms of the ecosystem supporting the initiative, as well as knowing which real-world packages we can use/reference when building additional functionality into the support tool.

We hope to see the ecosystem start to provide this information and look forward to seeing what tooling people (including the package-maintenance working group and others) come up with to help achieve the goals outlined.

Expedia Group: Building better testing pipelines with opensource.

By Blog, Case Study, ESLint, OpenJS In Action

The OpenJS In Action series features companies that use OpenJS Foundation projects to help develop efficient, effective web technologies. 

Software developers at global travel company Expedia Group are using JavaScript, ESLint and robust testing pipelines to reduce inconsistency and duplication in their code. Switching from Java and JSP to Node.js has streamlined development and design systems. Beyond that, Expedia developers are looking into creating a library of reusable design and data components for use across their many brands and pages. 

Robin Ginn, executive director of the OpenJS Foundation, interviewed Tiffany Le-Nguyen, Software Development Engineer at Expedia Group.

Expedia is an example of how adoption of new technologies and techniques can improve customer and developer experiences. 

A video featuring Expedia is available here:

Robin Ginn, executive director of the OpenJS Foundation, interviewed Tiffany Le-Nguyen, Software Development Engineer at Expedia Group. Le-Nguyen explained how accessibility and performance concerns led developers to modernize Expedia’s infrastructure. One of the choices they made was to integrate ESLint into their testing pipeline to catch bugs and format input before content was pushed live. ESLint also proved to be a huge time-saver — it enforced development standards and warned developers when incorrect practices were being used. 

ESLint was especially useful for guiding new developers through JavaScript, Node.js and TypeScript. Expedia made the bold move to switch most of their applications from Java and JSP to Node.js and TypeScript. Le-Nguyen is now able to catch most errors and quickly push out new features by combining Node.js with Express and a robust testing pipeline. 

However, Expedia is used globally to book properties and dates for trips. Users reserve properties with different currencies across different time zones. This makes it difficult to track when a property was reserved and whether the correct amount was paid. Luckily, Expedia was able to utilize Globalize, an OpenJS project that provides number formatting and parsing, date and time formatting and currency formatting for languages across the world. Le-Nguyen was able to simplify currency tracking across continents by integrating Globalize into the project. 

To end the talk, Le-Nguyen suggested that web developers should take another look into UI testing. Modern testing tools have simplified the previously clunky process. Proper implementation of a good testing pipeline improves the developer experience and leads to a better end product for the user. 

Ajv Joins OpenJS Foundation as an Incubation Project

By Announcement, Blog, Project Update

Today, Ajv, a JSON Schema validator for both server-side and client-side JavaScript applications, has entered into public incubation at the OpenJS Foundation. Ajv is a key part of the JavaScript ecosystem, used by a large number of web developers with millions of projects depending on it indirectly, via other libraries. 

In addition to becoming an incubating project, Ajv was recently awarded a grant from Mozilla’s Open Source Support (MOSS) program in the “Foundational Technology” track. This grant is continued validation for the important role Ajv plays within the JavaScript ecosystem and will help ensure this work continues. 

“A diverse set of widely used open source projects is why we exist and how our community continues to thrive,” said Robin Ginn, OpenJS Foundation Executive Director. “It’s great when these projects recognize the value of being part of the OpenJS community and benefit from what we are creating here. I’m thrilled to welcome Ajv as an incubation project to the OpenJS Foundation and excited to support its open development among web developers.”

Ajv is a leading JSON Schema validator that is highly specification compliant, supporting JSON Schema drafts 4 to 7. Ajv is also extensible via custom keywords and plugins, and is one of the fastest JSON Schema validators. Additionally, Ajv gets 120 million monthly downloads on npm. Many projects within the OpenJS Foundation use Ajv including webpack, ESlint, and Fastify.

“As CPC chair, I’m really happy that Ajv has become an incubating project at the OpenJS Foundation,” said Joe Sepi, OpenJS Foundation Cross Project Council Chair. “Ajv is an important project within the JavaScript open source space — many of our own projects already use it. This is an important step for Ajv and I, along with the entire CPC, am excited Ajv is taking this step with the OpenJS Foundation.”

“As Ajv’s CPC liaison, the person who helps guide potential projects through the application process, I’m excited for what’s to come for Ajv’s within the OpenJS Foundation,” said Dylan Schiemann, CEO at Living Spec and co-creator of Dojo. “As an incubating project, AJV has a unique opportunity to continue its path toward sustainability and growth. As a user of AJV and an early advocate for JSON Schema, we’re super excited to work with the project and support its growth as part of the OpenJS Foundation.”

“Ajv has become a centerpiece of all data-validation logic in my open-source projects and businesses. It’s spec-compliant, extensible, fast and has amazing support. Ajv joining the OpenJS Foundation will greatly benefit the entire JavaScript ecosystem,” said Gajus Kuizinas, CTO of Contra.

“I’ve been developing Ajv since 2015 and it is nice to see it being so widely used – it would never have happened without almost 100 contributors and a much larger number of users. Both the OpenJS Foundation and Mozilla grant will help Ajv become a permanent fixture in the JavaScript ecosystem – I am really looking forward to the next phase of Ajv development,” said Evgeny Poberezkin, the developer of Ajv.

By joining the OpenJS Foundation, there are multiple organizational and infrastructure areas that will be better supported. Furthermore, Ajv will be able to ensure governance and Code of Conduct enforcement to make sure that Ajv will continue to be stable. Joining will also help Ajv to grow and gain contributors, and potentially help with wider enterprise adoption through greater confidence and overall stability for the project.

As a collaborative project with transparency-first principles, the OpenJS Foundation is happy to welcome Ajv as an incubation project and looks forward to the many successes the project will have within its new home.

Start Contributing Now!

If you’d like to help build Ajv, you can start by looking at the Contributing Guidelines. Documentation, Issues, Bug Reports and lots more can be found here. Every contribution is appreciated! Please feel free to ask questions via Gitter chat.

AMA Recap from the Node.js Technical Steering Committee

By AMA, Blog, Node.js

Members of the Technical Steering Committee (TSC) for Node.js gave an informative AMA, which you can watch below. Speakers include Michael Dawson (@mhdawson1), Matteo Collina (@matteocollina), Gireesh Punathil (@gireeshpunam), Gabriel Schulhof (@gabrielschulhof), Bethany Griggs (@BethGriggs_), Colin Ihrig (@cjihrig), and Myles Borins (@MylesBorins).

Full video here

In this AMA, the TSC took questions from the live chat and gave insight into how they got involved. Questions ranged from whether Node.js is good for image processing to thoughts on Deno. The TSC focused on a mix of preexisting and user generated questions.

Beginning with suggestions on how to get involved with Node and ending on the same note, this AMA can inspire individuals to join Node.js.

Video by Section

Introductions (1:08)

How to Get Involved (4:48)

When To Update Your LTS? (13:45)

Is Node Good For Image Processing Applications? (34:45)

Upcoming 14RX (42:00)

What Do You Think About Deno? (44:20)

Yarn v2 Module (51:07)

Wrap Up (53:55)

Photo Credit: Myles Borins

Our next AMA will feature OpenJS Project NodeRED! Submit your questions for the NodeRED team here!

Node.js in a Kubernetes world

By Blog, Node.js, Node+JS Interactive, tutorial

7 basic tasks Node.js developers need to understand about Kubernetes

This post was written by Michael Dawson, OpenJS Foundation Board Member and Node.js community lead at IBM. This first appeared on IBM Developer.

Kubernetes is fast becoming the leader for deploying and managing production applications, including those written in Node.js. Kubernetes is complex, though, and learning the ins and outs of the technology can be difficult, even for a seasoned developer.

Kubernetes logo

Node.js application developers may not need to manage Kubernetes deployments in our day-to-day jobs or be experts in the technology, but we must consider Kubernetes when developing applications.

As a reminder, Docker and Kubernetes are the foundation of most modern clouds, including IBM CloudCloud native development refers to creating applications that you will deploy in Docker or Kubernetes, and cloud-native applications often conform to the tenants of a 12 Factor application.

Kubernetes logo

In this article, I try to answer the question: “As a Node.js devleoper, what do I need to know about Kubernetes?” I cover a number of key tasks you must complete when building your Node.js application in a cloud-native manner, including:

  • Building Docker images
  • Deploying containerized applications
  • Testing your applications
  • Health checking
  • Logging
  • Gathering metrics
  • Upgrading and maintaining containerized applications

Building Docker images

Deploying to Docker or Kubernetes requires that you build a Docker image. To do that, you typically start with an existing image and layer in the additional components that you need.

When building Node.js applications, the community provides a number of official Docker images that you can start with. These images add the Node.js binaries to an existing Linux distribution (Debian or Alpine) and offer a way to bundle your application into the image so that it runs when you start the Docker container.

Each time the project publishes a new release, it creates new Docker images. For example, when 12.13.1 was released, the new Docker image “node:12.13.1” was made available.

There are three variants of the Node.js images:

  1. Debian-based images with the core components needed to build and test Node.js applications.
  2. Slim images which are Debian-based images with only the minimal packages needed to run a Node.js application after it is already built.
  3. Alpine-based images for those who need the smallest container size.

There are a number of tags that you can use to specify the image you want. See Docker Hub for the full list.

When choosing your base image, keep in mind:

  • Alpine binaries are not built as part of the main Node.js releases. This means that they are built from source and there is more risk of regressions.
  • You should likely use both the standard Debian image along with the Slim image in a multistage build in order to achieve a smaller resulting image size. More on that in the next section

Build your image

Once you’ve chosen your base image, the next step is to actually build the image which bundles in your application. This is done through a Dockerfile. The simplest Dockerfile is as follows:

FROM node : 12.13.1
COPY server.js .
CMD node server.js

This Dockerfile simply copies your application code (server.js) into the official image with the node:12.13.1 binaries and indicates that server.js should be run when the image is started. Note that you did not have to install Node.js as it is already installed and in the default path as part of the node:12.13.1 base image.

To build an image from this file, put the file into a directory with your application code and then from that directory run docker build . -t test1:new. This builds a docker image named test1, and with the new tag, based on the contents of the current directory.

While this sounds straightforward, Dockerfiles can quickly become complicated because you want to build the image in a way that only includes the minimum components needed to run your application. This should exclude intermediate artifacts, for example the individual object files created when native addons are compiled. To exclude intermediate artifacts, a multistage build is commonly used where artifacts are generated in one step or image and then copied out to another image in a later stage. Read more about this in the Docker documentation.

Once you have built the image, you need to push it to a registry and then you can run and test your application under Docker or Kubernetes. It’s best to test in as close an environment to the production deployment as possible. Today, that often means Kubernetes.

Deploying your application to Kubernetes

In order to deploy and test your image in Kubernetes you use one or more deployment artifacts. At the lowest level, much of the Kubernetes configuration is through YAML. To deploy and access your application through a port, you need three YAML files.

To start, you need a deployment which specifies which image to use (the one you pushed to the registry earlier). For example:

apiVersion: apps/v1
kind: Deployment
    name: test1
        app: test1
    replicas: 2
            app: test1
                app: test1
            - name: test1
               image: test1:new
               imagePullPolicy: Never

In this case, it’s specifying through the image: entry that it should use the Docker image named test1 with the new tag (which matches what we built in the earlier step). You would use the following command to deploy to Kubernetes: kubectl apply -f deployment.yaml.

Similarly, you’d need to deploy a service to provide a way to get to the running containers that make up the deployment and an ingress to allow external access to that service:

kubectl apply -f service.yaml

kubectl apply -f ingress.yaml

I won’t go into the details here, but the key is that you need to understand and write YAML along with understanding the YAML for deployments, services, and ingress.

If this sounds a bit cumbersome, it is, and you are not the first one to think so. Helm charts were developed to bring together a deployment, service, and ingress in a way to simplify deployments.

The basic structure of a Helm chart is as follows:


The base Helm chart has templates which you can configure to avoid having to write your own deployment, service, and ingress YAML files. If you have not already written those files, it is probably easier to use that configuration option. You can, however, just copy your files into the template directory and then deploy the application with helm install dirname, where dirname is the name of the directory in which the chart files are located.

Helm is billed as the “Package Mana ger for Helm” and does make it easier to deploy an application configured through multiple YAML files and possibly multiple components. There are, however, shortcomings, including concerns over security and relying on only static configuration versus being able to have programmatic control. Operators look to address these issues. Read more about them here:

Iterative testing

After you figure out how to deploy your application, you need to test your application. The development and test process is iterative, so you probably need to rebuild, deploy, test a number of times. Check out our article on test-driven development.

If you are testing in Docker, you need to build a new image, start a container, and then access the image during each iteration. If you are testing in Kubernetes, you have to delete pods, redeploy, and so forth. Even when using Helm charts, this will take a number of commands and time for things to spin up. The key challenge is that this all takes time and that can slow down your development process.

So far, we’ve talked about building and testing an image for your application, but not what’s in your application itself. I can’t talk about your specific application, but let’s look at what you should build into your application in order to support a cloud-native deployment.

Health checking your applications

The first thing you need to build into your application is support for liveness and readiness endpoints. Kubernetes has built-in functionality to check these endpoints:

  • Liveness – restart the container when the liveness endpoint does not respond indicating the application is alive.
  • Readiness – defer sending traffic until application is ready to accept traffic, or step sending traffic if application is no longer able to accept traffic.

Kubernetes supports three probe types.

  • Running a command (through shell in the container)
  • HTTP probe
  • TCP probe

As a Node.js developer, you will probably use an HTTP probe. A response code of >=200 and less than 400 means everything is okay.

You configure the probes through the deployment YAML. For example:

apiVersion: v1
kind: Pod
     name: test1
          apps: test1
     - name: test1
        image: test1:latest
        imagePullPolicy: Never
        - /server
             path: /live
             port: 3000
           initialDelaySeconds: 5
           periodSeconds: 10
              path: /live
              port: 3000
            initialDelaySeconds: 1
            periodSeconds: 10

In this code listings, I added a liveness prob on port 3000 using the /live path.


  • initialDelaySeconds controls how quickly to start checking (in order to allow enough time for the application to start up).
  • periodSeconds controls how often to check.
  • timeoutSeconds is 1 by default, so if your probe can take longer than that you’ll have to set that value as well.
  • failureThreshold controls how many times the probe needs to fail before the container will restart. You want to ensure that you don’t get into a loop where the application continues to restart and never has long enough to become ready to serve the probes.

Similarly, I added a readinessProbe on port 3000 using the /ready path. To test this out, I created an application that only responds successfully to the readinessProbe after 10 seconds. If I query the pods after deploying the application the output is:

User10minikube1:~/test$ kubectl get all
NAME           READY       STATUS        RESTARTS        AGE
pod/test1      0/1         Running       0               6s

Which shows 0/1 under READY indicating that the application is not yet ready to accept traffic. After 10 seconds, it should show 1/1, and Kubernetes will start routing traffic to the pod. This is particularly important when we scale up or down as we don’t want to route traffic to a new container until it is ready or else it will slow responses even though another container is up and already capable of responding quickly.

The readiness and liveness probes may be as simple as:

app.get('/readiness', (req, res) => {



app.get('/liveness', (req, res) => {



However, you will often need to check the liveness and readiness of other components within your application. This can get complicated if you have a lot of subcomponents. If you have a lot of subcomponents to check, I recommend using a package like CloudNativeJS/cloud-health that allows you to register additional checks.


Node.js developers also need to know how to do logging in a cloud-native environment. In container development, writing logs out to disk does not generally make sense because of the extra steps needed to make the logs available outside the container—and they will be lost once the container is stopped.

Logging to standard out (stdout) is the Cloud native way, and structured logging (for example using JSON) is the current trend. One of my current favorite modules is pino which is a fast, structured logger and is easy to use.

Kubernetes has support for watching standard out, so it’s easy to get the log output, for example you can use kubectl logs -f <name of pod> where <name of pod> is the name of a specific pod as follows:

user1@minikube1:~/test$ kubectl get pods
NAME                     READY   STATUS    RESTARTS   AGE
test1-7bbc4449b4-98w2j   1/1     Running   0          30s
test1-7bbc4449b4-zqx6f   1/1     Running   0          30s
user1@minikube1:~/test$ kubectl logs -f test1-7bbc4449b4-98w2j

Cloud deployment environments like the IBM Cloud also have more sophisticated options for collecting and aggregating logs across containers, including LogDNA or the elastic stack Elasticsearch, Logstash, and Kibana.

Gathering metrics

In a Kubernetes deployment your application is running in containers, there may be multiple copies of each container, and it is not necessarily easy to find or access those containers in order to gather information about how your application is running. For this reason, it is important to export the key metrics from your container that you need to understand and track the health of your application.

Prometheus is the defacto standard on this front. It defines a set of metrics that you should export and gives you the ability to add additional application-specific metrics that are important to your business.

As an example, the following shows an example of the metrics returned using one of the available Prometheus packages:

# HELP process_cpu_user_seconds_total Total user CPU time spent in seconds.
# TYPE process_cpu_user_seconds_total counter
process_cpu_user_seconds_total 1.7980069999999984 1585160007722

# HELP process_cpu_system_seconds_total Total system CPU time spent in seconds.
# TYPE process_cpu_system_seconds_total counter
process_cpu_system_seconds_total 0.931571000000001 1585160007722

# HELP process_cpu_seconds_total Total user and system CPU time spent in seconds.
# TYPE process_cpu_seconds_total counter
process_cpu_seconds_total 2.7295780000000067 1585160007722

# HELP process_start_time_seconds Start time of the process since unix epoch in seconds.
# TYPE process_start_time_seconds gauge
process_start_time_seconds 1584451714

# HELP process_resident_memory_bytes Resident memory size in bytes.
# TYPE process_resident_memory_bytes gauge
process_resident_memory_bytes 17645568 1585160007723

# HELP process_virtual_memory_bytes Virtual memory size in bytes.
# TYPE process_virtual_memory_bytes gauge
process_virtual_memory_bytes 881635328 1585160007723

# HELP process_heap_bytes Process heap size in bytes.
# TYPE process_heap_bytes gauge
process_heap_bytes 90226688 1585160007723

# HELP process_open_fds Number of open file descriptors.
# TYPE process_open_fds gauge
process_open_fds 20 1585160007723

# HELP process_max_fds Maximum number of open file descriptors.
# TYPE process_max_fds gauge
process_max_fds 201354

# HELP nodejs_eventloop_lag_seconds Lag of event loop in seconds.
# TYPE nodejs_eventloop_lag_seconds gauge
nodejs_eventloop_lag_seconds 0.000200848 1585160007723

# HELP nodejs_active_handles Number of active libuv handles grouped by handle type. Every handle type is C++ class name.
# TYPE nodejs_active_handles gauge
nodejs_active_handles{type="Socket"} 2 1585160007722
nodejs_active_handles{type="Server"} 1 1585160007722

# HELP nodejs_active_handles_total Total number of active handles.
# TYPE nodejs_active_handles_total gauge
nodejs_active_handles_total 3 1585160007722

# HELP nodejs_active_requests Number of active libuv requests grouped by request type. Every request type is C++ class name.
# TYPE nodejs_active_requests gauge
nodejs_active_requests{type="FSReqCallback"} 2

# HELP nodejs_active_requests_total Total number of active requests.
# TYPE nodejs_active_requests_total gauge
nodejs_active_requests_total 2 1585160007722

# HELP nodejs_heap_size_total_bytes Process heap size from node.js in bytes.
# TYPE nodejs_heap_size_total_bytes gauge
nodejs_heap_size_total_bytes 5971968 1585160007723

# HELP nodejs_heap_size_used_bytes Process heap size used from node.js in bytes.
# TYPE nodejs_heap_size_used_bytes gauge
nodejs_heap_size_used_bytes 4394216 1585160007723

# HELP nodejs_external_memory_bytes Nodejs external memory size in bytes.
# TYPE nodejs_external_memory_bytes gauge
nodejs_external_memory_bytes 3036410 1585160007723

# HELP nodejs_heap_space_size_total_bytes Process heap space size total from node.js in bytes.
# TYPE nodejs_heap_space_size_total_bytes gauge
nodejs_heap_space_size_total_bytes{space="read_only"} 262144 1585160007723
nodejs_heap_space_size_total_bytes{space="new"} 1048576 1585160007723
nodejs_heap_space_size_total_bytes{space="old"} 3256320 1585160007723
nodejs_heap_space_size_total_bytes{space="code"} 425984 1585160007723
nodejs_heap_space_size_total_bytes{space="map"} 528384 1585160007723
nodejs_heap_space_size_total_bytes{space="large_object"} 401408 1585160007723
nodejs_heap_space_size_total_bytes{space="code_large_object"} 49152 1585160007723
nodejs_heap_space_size_total_bytes{space="new_large_object"} 0 1585160007723

# HELP nodejs_heap_space_size_used_bytes Process heap space size used from node.js in bytes.
# TYPE nodejs_heap_space_size_used_bytes gauge
nodejs_heap_space_size_used_bytes{space="read_only"} 32296 1585160007723
nodejs_heap_space_size_used_bytes{space="new"} 419112 1585160007723
nodejs_heap_space_size_used_bytes{space="old"} 2852360 1585160007723
nodejs_heap_space_size_used_bytes{space="code"} 386368 1585160007723
nodejs_heap_space_size_used_bytes{space="map"} 308800 1585160007723
nodejs_heap_space_size_used_bytes{space="large_object"} 393272 1585160007723
nodejs_heap_space_size_used_bytes{space="code_large_object"} 3552 1585160007723
nodejs_heap_space_size_used_bytes{space="new_large_object"} 0 1585160007723

# HELP nodejs_heap_space_size_available_bytes Process heap space size available from node.js in bytes.
# TYPE nodejs_heap_space_size_available_bytes gauge
nodejs_heap_space_size_available_bytes{space="read_only"} 229576 1585160007723
nodejs_heap_space_size_available_bytes{space="new"} 628376 1585160007723
nodejs_heap_space_size_available_bytes{space="old"} 319880 1585160007723
nodejs_heap_space_size_available_bytes{space="code"} 3104 1585160007723
nodejs_heap_space_size_available_bytes{space="map"} 217168 1585160007723
nodejs_heap_space_size_available_bytes{space="large_object"} 0 1585160007723
nodejs_heap_space_size_available_bytes{space="code_large_object"} 0 1585160007723
nodejs_heap_space_size_available_bytes{space="new_large_object"} 1047488 1585160007723

# HELP nodejs_version_info Node.js version info.
# TYPE nodejs_version_info gauge
nodejs_version_info{version="v12.13.1",major="12",minor="13",patch="1"} 1

There are Node.js packages that can help you export the recommended set of metrics and the ones you need that are specific to your application. prom-client is my current favorite for integrating Prometheus metrics into a Node.js application.

Kubernetes also generates Prometheus metrics, and Kubernetes distributions or cloud platforms often make it easy to scrape your requested endpoints, aggregate them, and then graph the resulting data. Even if you install Kubernetes locally, there are Helm charts that make it easy to do this. The end result is that if you export Prometheus metrics from your application it will generally be quite easy for you or your operations teams to aggregate, chart, and alert on this data.

Maintaining and upgrading your applications

As with all application development, you have to think about maintaining and upgrading your application. Kubernetes helps you update your application deployment by rolling out updated containers, but it does not help you update the contents of the containers you built.

You have to have a plan for how to maintain and upgrade each component you use to build your container or that you add into your container. These range from docker images, Dockerfiles, live and readiness endpoints, logging, and metrics.

If you have X projects with Y containers, you can end up with a big number of container images that you must maintain. When you decide to update a dependency (for example the logger or Prometheus package etc.), you have to figure out which images need to be updated, what versions they are already using, and the like.

What happens if you move to a new team? You’ll need to figure out what their docker file look like, how the deployment artifacts are organized, what endpoint names are used for liveness and readiness, which logger they are using, and more just to get started.

I don’t know about you, but at this point I start to think “enough is enough” — I just want to focus on the application code —. Do I really need to think about Docker, Dockerfiles, YAML and all this stuff that’s outside of my application itself?

Separating concerns

What I’d really like is some separation of concerns and tooling to help me build/test efficiently using best practices agreed on by my organization. Something along these lines:

Kubernetes output

The two main components are:

  • a common stack agreed on by developers, architects, and operations which is re-used across projects within my organization
  • the application and additional packages that I need for the applications that I’m working on

For me, having consistency in the areas that don’t affect the “secret sauce” for the application I’m working on brings great benefits in ongoing maintenance and upgrades and even moving across teams. I look at this as Consistency where it makes sense as opposed to uniformity.

Tools to help with consistency

There are a few tools that can help isolate the tasks that you don’t need to worry about so that you can just focus on your code. I’ll introduce three new ones that I think can help you succeed.


The Appsody open source project attempts to separate the operations aspects of application development to allow developers to focus on their application code, while at the same time building on a common Stack(s) that can be shared across an organization.

Appsody helps in the following ways:

  • Builds a base stack that you can layer applications on top of. This include most of the base components I’ve mentioned above which are needed for cloud-native deployment.
  • Accelerates the iterative local development and test cycle.
  • Simplifies creating Docker images using best practices and deploying them to Kubernetes.

There are Node.js-specific stacks in Appsody, including

  • nodejs – base stack for any Node.js application
  • nodejs-express – base stack with express built in
  • nodejs-loopback – base stack with loopback built in
  • nodejs-functions – base stack where you can write functions in the Function as a service(FaaS) style

I’m excited about Appsody offering reusable stacks that can be shared across teams, and that allows architects and operators to update the stack separately from the application itself. This is a newish approach (hence, the incubator in the path for the stacks), so I won’t be surprised if there are tweaks to the approach or tools as people experiment and get more familiar with it.

It’s great to see an open source code base starting to form and allow organizations to start proving out the approach. The existing stacks integrate liveness, readiness endpoints, and export a Prometheus endpoint. At the same time, the base Appsody framework also provides commands for building and deploying containers to Kubernetes or testing locally in Docker. The result is that much of what you need to know or understand as a Node.js developer for a coud-native deployment is supported by the tooling.

While I still think it is good for you to have a basic understanding of Kubernetes and cloud-native development, these new tools ensure you won’t need to be an expert on the ins and outs of the deployment YAML files, multistage Docker files, and the like.


If you like what you see in Appsody but prefer a graphical user interface, you might want to check out the open source Codewind project. It adds a nice GUI, has Appsody integration, and allows you to bring your own visual editor (for example VS Code). It also adds in performance monitoring and Tekton pipelines. I’m more of a command line vi person, but if you’re interested you can check out it out here.

IBM Cloud Pack for Applications

Finally, if you are sold on this approach and want to build on a supported stack from IBM, check out the IBM Cloud Pack for Applications which includes the stack- based approach as well as support for Node.js.

Final thoughts

I hope you’ve gained a more concrete understanding of what cloud-native development and Kubernetes means to you as a Node.js developer and hope I’ve piqued your interest in stack-based development and Appsody. I talked on this subject at the most recent Node+JS Interactive conference and if you’d like to watch the recording, check out the slides and watch the recording.

Controlling Appium via raw HTTP requests with curl

By Appium, Blog, tutorial

This post originally appeared in Appium::Pro, Appium’s newsletter. Appium, a hosted Project at the OpenJS Foundation, is an open-source platform that enables automated testing of mobile and desktop apps on iOS, Android, Windows, Mac, and more.

Did you know that Appium is just a web server? That’s right! I’m going to write a full edition at some point on the WebDriver Protocol, which is the API that Appium implements (and which, as you can tell by the name, was invented for the purpose of browser automation). But for now, let’s revel in the knowledge that Appium is just like your backend web or API server at your company. In fact, you could host Appium on a server connected to the public internet, and give anybody the chance to run sessions on your devices! (Don’t do this, of course, unless you’re a cloud Appium vendor).

You may be used to writing Appium code in Java or some other language, that looks like this:

driver = new IOSDriver<WebElement>(new URL("http://localhost:4723/wd/hub"), capabilities);
WebElement el = driver.findElement(locator);

This looks like Java code, but what’s happening under the hood is that each of these commands is actually triggering an HTTP request to the Appium server (that’s why we need to specify the location of the Appium server on the network, in the IOSDriver constructor). We could have written all the same code, for example, in Python:

driver = webdriver.Remote("http://localhost:4723/wd/hub", capabilities)
el = driver.find_element(locator)

In both of these cases, while the surface code looks different, the underlying HTTP requests sent to the Appium server (and the responses coming back) are the same! This is what allows Appium (and Selenium, where we stole this architecture from) to work in any programming language. All someone needs to do is code up a nice little client library for that language, that converts that language’s constructs to HTTP requests.

What all this means is that we technically don’t need a client library at all. It’s convenient to use one, absolutely. But sometimes, we want to just run an ad-hoc command against the Appium server, and creating a whole new code file and trying to remember all the appropriate syntax might be too much work. In this case, we can just use curl, which is a command line tool used for constructing HTTP requests and showing HTTP responses. Curl works on any platform, so you can download it for your environment if it’s not there already (it comes by default on Macs, for example). There are lots of options for using curl, and to use it successfully on your own, you should understand all the components of HTTP requests. But for now, let’s take a look at how we might encode the previous four commands, without any Appium client at all, just by using curl!

# 0. Check the Appium server is online
> curl http://localhost:4723/wd/hub/status

# response:
# 1. Create a new session
> curl -H 'Content-type: application/json' \
http://localhost:4723/wd/hub/session \
-d '{"capabilities": {"alwaysMatch": {"platformName": "iOS", "platformVersion": "13.3", "browserName": "Safari", "deviceName": "iPhone 11"}}}'

# response:
{"value":{"capabilities":{"webStorageEnabled":false,"locationContextEnabled":false,"browserName":"Safari","platform":"MAC","javascriptEnabled":true,"databaseEnabled":false,"takesScreenshot":true,"networkConnectionEnabled":false,"platformName":"iOS","platformVersion":"13.3","deviceName":"iPhone 11","udid":"140472E9-8733-44FD-B8A1-CDCFF51BD071"},"sessionId":"ac3dbaf9-3b4e-43a2-9416-1a207cdf52da"}}

# save session id
> export sid="ac3dbaf9-3b4e-43a2-9416-1a207cdf52da"

Let’s break this one down line by line:

  1. Here we invoke the curl command, passing the -H flag in order to set an HTTP request header. The header we set is the Content-type header, with value application/json. This is so the Appium server knows we are sending a JSON string as the body of the request. Why do we need to send a body? Because we have to tell Appium what we want to automate (our “capabilities”)!
  2. -X POST tells curl we want to make a POST request. We’re making a POST request because the WebDriver spec defines the new session creation command in a way which expects a POST request.
  3. We need to include our URL, which in this case is the base URL of the Appium server, plus /session because that is the route defined for creating a new session.
  4. Finally, we need to include our capabilities. This is achieved by specifying a POST body with the -d flag. Then, we wrap up our capabilities as a JSON object inside of an alwaysMatch and a capabilities key.

Running this command, I see my simulator pop up and a session launch with Safari. (Did the session go away before you have time to do anything else? Then make sure you set the newCommandTimeout capability to 0). We also get a bunch of output like in the block above. This is the result of the new session command. The thing I care most about here is the sessionId value of ac3dbaf9-3b4e-43a2-9416-1a207cdf52da, because I will need this to make future requests! Remember that HTTP requests are stateless, so for us to keep sending automation commands to the correct device, we need to include the session ID for subsequent commands, so that Appium knows where to direct each command. To save it, I can just export it as the $sid shell variable.

Now, let’s find an element! There’s just one element in Appium’s little Safari welcome page, so we can find it by its tag name:

# 2. Find an element
> curl -H 'Content-type: application/json' \
-X POST http://localhost:4723/wd/hub/session/$sid/element \
-d '{"using": "tag name", "value": "h1"}'

# response:

# save element id:
> export eid="5000"

In the curl command above, we’re making another POST request, but this time to /wd/hub/session/$sid/element. Note the use of the $sid variable here, so that we can target the running session. This route is the one we need to hit in order to find an element. When finding an element with Appium, two parameters are required: a locator strategy (in our case, “tag name”) and a selector (in our case, “h1”). The API is designed such that the locator strategy parameter is called using and the selector parameter is called value, so that is what we have to include in the JSON body.

The response we get back is itself a JSON object, whose value consists of two keys. The reason there are two keys here is a bit complicated, but what matters is that they each convey the same information, namely the ID of the element which was just found by our search (5000). Just like we did with the session ID, we can store the element ID for use in future commands. Speaking of future commands, let’s get the text of this element!

# 3. Get text of an element
> curl http://localhost:4723/wd/hub/session/$sid/element/$eid/text

# response:
{"value":"Let's browse!"}

This curl command is quite a bit simpler, because retrieving the text of an element is a GET command to the endpoint /session/$sid/element/$eid/text, and we don’t need any additional parameters. Notice how here we are using both the session ID and the element ID, so that Appium knows which session and which element we’re referring to (because, again, we might have multiple sessions running, or multiple elements that we’ve found in a particular session). The response value is the text of the element, which is exactly what we were hoping to find! Now all that’s left is to clean up our session:

# 4. Quit the session
> curl -X DELETE http://localhost:4723/wd/hub/session/$sid

# response:

This last command can use all the default curl arguments except we need to specify that we are making a DELETE request, since that is what the WebDriver protocol requires for ending a session. We make this request to the endpoint /session/$sid, which includes our session ID so Appium knows which session to shut down.

That’s it! I hope you’ve enjoyed learning how to achieve some “low level” HTTP-based control over your Appium (and Selenium) servers!

Project News: WebdriverIO ships v6

By Announcement, Blog, Project Updates, WebdriverIO

Kudos to the WebdriverIO team for their recent v 6 release. Webdriver, a hosted project at the OpenJS Foundation, is a Next-gen browser automation test framework for Node.js

Big updates include:

Drop Node v8 Support
WebDriver has dropped support for Node v8, which was deprecated by the Node.js team at the start of 2020. It is not recommended to run any systems using that version anymore. It is strongly advised to switch to Node v12 which will be supported until April 2022.

Automation Protocol is now Default
Because of the great success of automation tools like Puppeteer and it became obvious that the WebDriver protocol in its current shape and form doesn’t meet the requirements of today’s developer and automation engineers. Members of the WebdriverIO project are part of the W3C Working Group that defines the WebDriver specification and they work together with browser vendors on solutions to improve the current state of the art. Thanks to folks from Microsoft there already proposals about a new bidirectional connection similar to other automation protocols like Chrome Devtools.

Performance Improvements
A big goal with the new release was to make WebdriverIO more performant and faster. Running tests on Puppeteer can already speed up local execution. Additionally, v6 has replaced the heavy dependency to request which has been fully depreciated as of February 11th 2020. With that, the bundle size of the webdriver and webdriverio package has been decreased by 4x.

These are only a few things that the v6 release brings. Read the full blog on the WebdriverIO site