The information in our interactive Radar is currently only available in English. To get information in your native language, please download the PDF here.


  • Since we last mentioned Kubernetes in the Radar, it has become the default solution for most of our clients when deploying containers into a cluster of machines. The alternatives didn’t capture as much mindshare, and in some cases our clients are even changing their ‘engine’ to Kubernetes. Kubernetes has become the container orchestration platform of choice for major public cloud platforms, including Microsoft's Azure Container Service and Google Cloud (see the GKE blip). And there are many useful products enriching the fast-growing Kubernetes ecosystem. Platforms that try to hide Kubernetes under an abstraction layer, however, have yet to prove themselves.


  • We're seeing increased adoption of .NET Core, the open source cross-platform software framework. .NET Core enables the development and deployment of .NET applications on Windows, macOS and Linux. With the release of .NET Standard 2.0 increasing the number of standard APIs across .NET platforms, the migration path to .NET Core has become clearer. Issues related to library support on .NET Core are becoming less problematic, and first-class cross-platform tooling is now available, allowing for productive development on non-Windows platforms. Blessed Docker images are provided to make it easy to integrate .NET Core services into a containerized environment. Positive directions in the community and feedback from our projects indicate that .NET Core is ready for widespread use.

  • The huge number of mobile devices makes it almost impossible for companies to test their mobile apps on all of them. Enter AWS Device Farm, an app-testing service that enables you to run and interact with your Android, iOS and web apps on a wide variety of physical devices that are hosted in the cloud simultaneously. Detailed logs, performance graphs and screenshots are generated during each run to provide general and device-specific feedback. The service offers a lot of flexibility by allowing the state and configuration of each device to be altered in order to reproduce very specific test scenarios. Our teams are using AWS Device Farm to run end-to-end tests on devices with the largest install base for their apps.

  • Load testing became easier with the maturity of tools such as Gatling and Locust. At the same time, elastic cloud infrastructures make it possible to simulate a large number of client instances. We're delighted to see Flood and other cloud platforms go further by leveraging these technologies. Flood IO is an SaaS load-testing service that helps to distribute and execute testing scripts across hundreds of servers in the cloud. Our teams find it simple to migrate performance testing to Flood by reusing existing Gatling scripts.

  • As Google Cloud Platform (GCP) has expanded in terms of available geographic regions and maturity of services, customers globally can now seriously consider it for their cloud strategy. In some areas, GCP has reached feature parity with its main competitor, Amazon Web Services, while in other areas it has differentiated itself — notably with accessible machine learning platforms, data engineering tools, and a workable Kubernetes as a service solution (GKE). In practice, our teams have nothing but praise for the developer experience working with the GCP tools and APIs.

  • In a microservice, or any other distributed architecture, one of the most common needs is to secure the services or APIs through authentication and authorization features. This is where Keycloak comes in. Keycloak is an open source identity and access management solution that makes it easy to secure applications or microservices with little to no code. It supports single sign-on, social login and standard protocols such as OpenID Connect, OAuth 2.0 and SAML out of the box. Our teams have been using this tool and plan to keep using it for the foreseeable future. But it requires a little work to set up. Because configuration happens both at initialization and at runtime through APIs, it's necessary to write scripts to ensure deployments are repeatable.

  • As monolithic applications are being replaced with more complex (micro)service ecosystems, tracing requests across multiple services is becoming the norm. With majority contribution from LightStep and Uber OpenTracing is rapidly becoming the de facto standard for distributed tracing. There is a growing number of tracers supporting OpenTracing standard, including Zipkin, Instana, and Jaeger. OpenTracing currently provides vendor-neutral implementation in multiple languages including: Go, JavaScript, Java, Python, Objective-C, C#, C++, Ruby and PHP.

  • In previous Radars, we mentioned that Unity has become the platform of choice for VR and AR application development because it provides the abstractions and tooling of a mature platform, while being more accessible than its main alternative, the Unreal Engine. With the recent introductions of ARKit for iOS and ARCore for Android, the two main mobile platforms now have powerful native SDKs for building augmented reality applications. Yet, we feel that many teams, especially those without deep experience in building games, will benefit from using an abstraction such as Unity, which is why we're calling out Unity beyond gaming. This allows developers unfamiliar with the technology to focus on one SDK. It also offers a solution for the huge number of devices, especially on the Android side, that are not supported by the native SDKs.

  • WeChat, often seen as a WhatsApp equivalent, is becoming the de facto business platform in China. Many people may not know but WeChat is also one of the most popular online payment platforms. With the app's built-in CMS and membership management, small businesses are now conducting their commerce entirely on WeChat. Through the Service Account feature, large organizations can interface their internal system to their employees. Given that more than 70 percent of Chinese people are using WeChat, it's an important consideration for businesses that want to expand into the China market.


  • Azure Service Fabric is a distributed systems platform built for microservices and containers. It’s comparable to container orchestrators such as Kubernetes, but also works with plain old services. It can be used in a bewildering array of ways, starting from simple services in your language of choice to Docker containers or services built using an SDK. Since its release a couple of years ago, it has steadily added more features, including Linux container support. Kubernetes has been the poster child of container orchestration tools, but Service Fabric is the default choice for .NET applications. We're using it in a few projects at ThoughtWorks and we like what we’ve seen so far.

  • Cloud Spanner is a fully managed relational database service offering high availability and strong consistency without compromising latency. Google has been working on a globally distributed database called Spanner for quite some time. It has recently released the service to the outside world as Cloud Spanner. You can scale your database instance from one to thousands of nodes across the globe without worrying about data consistency. By levering TrueTime, a highly available and distributed clock, Cloud Spanner provides strong consistency for reads and snapshots. You can use standard SQL to read data from Cloud Spanner, but for write operations you have to use their RPC API. Although not all services would require a global-scale distributed database, the general availability of Cloud Spanner is a big shift in the way we think about databases. And its design is influencing open source products such as CockroachDB.

  • After thorough exploration, R3, an important player in the blockchain space, realized that blockchain doesn't fit their purpose well, so they created Corda. Corda is a distributed ledger technology (DLT) platform focused on the financial field. R3 have a very clear value proposition and know that their problem requires a pragmatic technology approach. This matches our own experience; current blockchain solutions may not be the reasonable choice for some business cases, due to mining costs and operational inefficiency. Although the development experience we have on Corda thus far has not been the smoothest, APIs are still unstable after v1.0 release, we expect to see the DLT space mature further.

  • Cosmos DB is Microsoft's globally distributed, multimodel database service, which became generally available earlier this year. While most modern NoSQL databases offer tunable consistency, Cosmos DB makes it a first-class citizen and offers five different consistency models. It's worth highlighting that it also supports multiple models — key value, document, column family and graph — all of which map to its internal data model, called atom-record-sequence (ARS). One interesting aspect of Cosmos DB is that it offers service level agreements (SLAs) on its latency, throughput, consistency and availability. With its wide range of applicability, it has set a high standard for other cloud vendors to match.

  • In parallel with the recent surge of chatbots and voice platforms, we've seen a proliferation of tools and platforms that provide a service to extract intent from text and management of conversational flows that you can hook into. DialogFlow (formerly, which was acquired by Google, is one such ‘natural-language-understanding as a service’ offering that competes with and Amazon Lex among other players in this space.

  • While the software development ecosystem is converging on Kubernetes as the orchestration platform for containers, running Kubernetes clusters remains operationally complex. GKE (Google Kubernetes Engine) is a managed Kubernetes solution for deploying containerized applications that alleviates the operational overhead of running and maintaining Kubernetes clusters. Our teams have had a good experience using GKE, with the platform doing the heavy lifting of applying security patches, monitoring and auto-repairing the nodes, and managing multicluster and multiregion networking. In our experience, Google's API-first approach in exposing platform capabilities, as well as using industry standards such as OAuth for service authorization, improve the developer experience. It's important to consider that GKE is under rapid development which, despite the developers' best efforts to abstract consumers from underlying changes, has impacted us temporarily in the past. We're expecting continuous improvement around maturity of Infrastructure as code with Terraform on GKE and similar tools.

  • Hyperledger is a platform built around blockchain technologies. It consists of a blockchain implementation named Fabric and other associated tools. Disregarding the hype surrounding blockchain, our teams have found it easy to get started with these tools. The fact that it is an open source platform supported by the Linux Foundation also adds to our excitement about Hyperledger.

  • Kafka Streams is a lightweight library for building streaming applications. It's been designed with the goal of simplifying stream processing enough to make it easily accessible as a mainstream application programming model for asynchronous services. It can be a good alternative in scenarios where you want to apply a stream processing model to your problem, without embracing the complexity of running a cluster (usually introduced by full-fledged stream processing frameworks). New developments include ‘exactly once’ stream processing in a Kafka cluster. This was achieved by introducing idempotency in Kafka producers and allowing atomic writes across multiple partitions using the new Transactions API.

  • Much of the power of sophisticated IDEs comes from their ability to parse a program into an abstract syntax tree (AST) and then use that AST for program analysis and manipulation. This supports features such as autocomplete, finding callers and refactoring. Language servers pull this capability into a process that allows any text editor to access an API to work with the AST. Microsoft has led the creation of the Language Server Protocol (LSP), harvested from their OmniSharp and TypeScript Server projects. Any editor that uses this protocol can work with any language that has an LSP-compliant server. This means we can keep using our favorite editors without forgoing the rich text editing modes of many languages — much to the delight of our Emacs addicts.

  • LoRaWAN is a low-power wide-area network, designed for low-power consumption and communication over long distances using low bitrates. It provides for communication between devices and gateways, which can then forward the data to, for example, applications or servers. A typical usage is for a distributed set of sensors, or for Internet of Things (IoT) devices, for which long battery life and long-range communication is a must. LoRaWAN addresses two of the key problems with attempting to use normal Wi-Fi for such applications: range and power consumption. There are several implementations, a notable one being The Things Network, a free, open source implementation.

  • MapD is an in-memory columnar analytic database with SQL support that's built to run on GPU. We debated whether the database workload is actually I/O or computationally bound but there are instances where the parallelism of the GPU, combined with the large bandwidth of VRAM, can be quite useful. MapD transparently manages the most frequently used data in VRAM (such as columns involved in group-by, filters, calculations and join conditions) and stores the rest of the data in the main memory. With this memory management setup, MapD achieves significant query performance without the need of indexes. Although there are other GPU database vendors, MapD is leading this segment with the recent open source release of its core database and through the GPU Open Analytics Initiative. If your analytical workload is computationally heavy, can exploit GPU parallelism and can fit in the main memory, we recommend assessing MapD.

  • In our experience—for Internet of Things (IoT) solutions where a lot of devices communicate with each other and/or a central data hub—the MQTT connectivity protocol has proven itself. We've also come to like the Mosquitto MQTT broker. It might not satisfy all demands, particularly with regard to scalability, but its compact nature and easy setup makes it ideal for development and testing purposes.

  • We like simple tools that solve one problem really well, and Netlify fits this description nicely. You can create static website content, check it into GitHub and then quickly and easily get your site live and available. There is a CLI available to control the process; content delivery networks (CDNs) are supported; it can work alongside tools such as Grunt; and, most importantly, Netlify supports HTTPS.

  • PlatformIO provides a rich ecosystem for IoT development by providing cross-platform builds, library management and good integration with existing IDEs. The intelligent code completion and Smart Code Linter with built-in terminal and serial port monitor greatly enhances the developer experience. It also organizes and maintains thousands of libraries and provides a clean dependency manager with semantic versioning to ease IoT development. We've started using PlatformIO in a few IoT projects and we really like it for its simplicity and wide support of platforms and boards.

  • Machine-learning models are starting to creep into everyday business applications. When enough training data is available, these algorithms can address problems that might have previously required complex statistical models or heuristics. As we move from experimental use to production, we need a reliable way to host and deploy the models that can be accessed remotely and scale with the number of consumers. TensorFlow Serving addresses part of that problem by exposing a remote gRPC interface to an exported model; this allows a trained model to be deployed in a variety of ways. TensorFlow Serving also accepts a stream of models to incorporate continuous training updates, and its authors maintain a Dockerfile to ease the deployment process. Presumably, the choice of gRPC is to be consistent with the TensorFlow execution model; however, we’re generally wary of protocols that require code generation and native bindings.

  • Voice platforms such as Amazon Alexa and Google Home are riding high on the hype cycle; some even herald the ubiquity of the conversational voice interface. We're already integrating conversational UIs into products and seeing the impact of this new interaction in how we design interfaces. Alexa specifically was built from the ground up without a screen and treats the conversational UI as first-class. But it's still too early to believe the hype, and we expect more big players to get in the game.

  • Microsoft is catching up in the container space with Windows Containers. At the time of writing, Microsoft provides two Windows OS images as Docker containers, Windows Server 2016 Server Core and Windows Server 2016 Nano Server. Although there is room for improvement for Windows Containers, for instance, decreasing the large image sizes, and enriching ecosystem support and documentation, our teams have started using them in scenarios where other containers have been working successfully, such as build agents.


  • We remain concerned about business logic and process orchestration implemented in middleware, especially where it requires expert skills and tooling while creating single points of scaling and control. Vendors in the highly competitive API gateway market are continuing this trend by adding features through which they attempt to differentiate their products. This results in overambitious API gateway products whose functionality — on top of what is essentially a reverse proxy — encourages designs that continue to be difficult to test and deploy. API gateways do provide utility in dealing with some specific concerns — such as authentication and rate limiting — but any domain smarts should live in applications or services.

Unable to find something you expected to see? Your item may have been on a previous radar