This article is contributed. See the original author and article here.
In case you hadn’t noticed, it’s a polyglot world. The number of scenarios where your application needs to communicate with services using a runtime you know nothing about is greater than ever. Fortunately, things like RESTful endpoints and JSON make life a lot easier. But what if your application needs to share settings with other applications? Or what if a backing service’s address changes between datacenters? How will your application know to switch?
The answer – as you may have guessed – doesn’t come right off the shelf. In this post we will look at some best practice solutions using an external configuration provider and a service registry. Both will complicate local development and testing, as well as add some debt to the person (or team) that maintains it in production. Both solutions have a ton of configurable possibilities, which, of course, makes things even more complicated.
Let’s focus on (arguably) the two most popular products to solve these challenges. As an external configuration provider, Spring Cloud Config is a top performer. The most popular way to configure the value store is with a Git based repository. Just check-in a YAML file of name-value pairs and tell Config Server how to connect with it. Config server will monitor the repo for future changes and interpret things like different profiles for different environments.
Using Eureka from the Spring Cloud Netflix project is a safe bet to implement service discovery. It’s a time tested solution that came from an organization that knows a little something about microservices. Starting a registry instance doesn’t have a whole lot of overhead. The work is more focused on each application that wants to register itself or discover register services. The app needs to implement a Eureka client for proper interaction, and it has lots of possibilities.
Config Server and Eureka Discovery themselves are worthy of their own discussion, which I’ll leave to folks smarter than me to cover. The fun really starts when you combine them both. Say, for example, you have an ASP.NET Core microservice that follows the config first pattern. That means in appsettings.json you add two simple values – the address of the config server and turn on FailFast. That’s it. The application won’t start if it can’t contact the server. Upon successfully connecting the application’s default logging levels, address to discovery server, management endpoint policies, connection strings, and everything else you can imagine will be loaded in.
The application also relies on the service discovery pattern to find the address of other services. For instance, instead of setting the HTTP client’s base URL to an IP address or FQDN, the application instead uses the name of the service like ‘https://another-service/api/something/2’ as the base. A request is first sent to the discovery service to find addresses where instances of the service can be reached. Then Steeltoe uses an ILoadBalancer to decide the best address to use. The decision could be based on a load balancing algorithm, the health of instances, or other strategies. Steeltoe then forwards the request to the actual service, and the response comes directly back to the calling application. To you (the Developer) all of this is wrapped up in a single HTTPClient call.
This combination of discovery and external config is quite powerful in situations where settings and service addresses stay fluid – i.e. the cloud. But I have yet to meet a developer that is willing to manage all this for the entire organization.
Meet the Pet Clinic application
In the Spring Java world, there is an application known as Pet Clinic that has been around since the dawn of…well… Spring. Its purpose is to highlight the many features of Spring and has become so popular that it has its own GitHub org. Of the many Pet Clinic derivations, the spring-petclinic-microservices version offers a relatively real-world cloud platform scenario. This version breaks the application into different microservices and uses Spring API Gateway to host the web UI as well as host a few public endpoints.
Note that API Gateway doesn’t have any backing database. Its endpoints dynamically create responses by calling other backing services and aggregating the results together. It insulates those backing services from ever having to be publicly available.
Referring to the below architecture drawing you’ll see the relationship Eureka and Config server play. Almost everything relies on retrieving config dynamically and either discovering service addresses or registering itself as discoverable.
Also of note is the Spring Boot Admin Server. This service packs a powerful debugging punch and can serve as a priceless resource to developers wanting to see exactly how each application’s container is configured and turn logging levels up and down instantly. Admin Server can provide so much insight into all the services because the ‘actuator’ feature is enabled in each. You can learn more about Spring Boot Actuators for Spring Java apps and Steeltoe Management Endpoints for .NET apps in their respective docs.
The last feature of the Pet Clinic Microservices architecture is for performance & debugging with distributed tracing. In this case, Zipkin is the chosen server. Each application is given direction to export traces to the Zipkin server in a compliant format. The power here is around simplicity. In Spring you use Sleuth and in Steeltoe you use the Zipkin exporter. Both are dead simple to implement and know exactly how to monitor requests and build spans into traces. In fact, both can also monitor an application’s logging and correlate logs with request traces! This is a very powerful feature for developing and debugging.
Deploy Pet Clinic on Azure Spring Cloud
As you review the previous architecture, the elephant in the room is how many working parts the Pet Clinic Microservice application has. Each service brings value and takes the application to a higher cloud-native status, but adds to the management burden.
Azure Spring Cloud aims to minimize the management burden. When you create an instance of Azure Spring Cloud you get a fully managed Eureka, Config, and Zipkin compatible tracing service. They have taken care of the hard stuff. Each application running in Azure Spring Cloud is given the address to these services in a secure manner. This means Spring and Steeltoe applications are going to feel right at home the second they start running.
The customers, vets, and visits services don’t need to be public, only the API Gateway. The Admin Server also doesn’t need a public address. Instead it can use the automatically generated “Test Endpoint” that is behind Azure auth.
The applications still export traces to a Zipkin compatible service. The service provides a very useful map to drill into details via Application Map in Azure Application Insights.
As you can see Azure Spring Cloud helps developers be more cloud-native by managing the infrastructure and providing native bindings to Spring and Steeltoe thus minimizing the maintenance burden.
Get Started
For .NET developers, the Steeltoe team has refactored each Spring microservice to ASP.NET Core 3.1 and added in all the Steeltoe fixin’s so everything just works. Review the source code here and get started. The main branch of the application is meant for local development. Note, to run the entire application, you’ll need maven to compile the Spring pieces and (within the Docker Compose) you’ll see the Dockerfile used for compiling the .NET services. On my Windows desktop I used a combination of Docker Desktop with WSL and VS Code with WSL to get everything running.
If you would rather run everything as Spring Boot (no .NET services), refer to the spring-petclinic-microservices repo. There you will find exact instructions to run the application locally.
Once you are ready to get started with Azure Spring Cloud, you have quite a few options. If you prefer to get a little more familiar with Azure Spring Cloud without diving in with Pet Clinic, you’ll find how to’s, tutorials, overview, and getting started guides for both Java and .NET in the documentation. There are step by step instructions to help get familiar with the overall service and deploy different flavors of microservices.
When you’re ready to see Azure Spring Cloud in full swing, head over to the azure-spring-cloud branch of the Steeltoe application. There you will find a “deploy-to-Azure Spring Cloud” script in both powershell and bash. Either will get everything going in Azure Spring Cloud and help you see the full potential of the Azure Spring Cloud managed service.
You can also find a Spring only version of Pet Clinic for Azure Spring Cloud in the Azure samples.
Brought to you by Dr. Ware, Microsoft Office 365 Silver Partner, Charleston SC.
Recent Comments