Service Discovery in a Microservices Architecture: Client vs Service side discovery

Reading Time: 11 minutes

Service discovery is one of the key things of distributed systems. This allow us to automatically discover services on a network. In order to make some REST request, your service needs to know the service location. This service location includes IP address and port of a service instance. In the most scenarios, from the same service will be deployed multiple instances. The traditional applications are running on a physical hardware and the network locations are static. On the other hand, in the modern cloud-based microservice solutions, this network locations are dynamic. This makes them much harder to manage and more difficult challenge.

There are two main service discovery patterns:

  • Client-side discovery (Eureka)
  • Service-side discovery (Kubernetes)

Service Registry?

The Service Registry is a process that involves registering a service’s location to some central place. You can imagine that as a database of a service locations. The Instances of services would be register into the service registry on startup and deregistered accordingly. It will register it’s host, ports and some additional information like on the image below.
I have already created example application using Netflix Eureka service registry. It comes with some of the predefined annotations. For this purpose, I have created two spring applications. One of application will act as a discovery server. The application needs to include the following dependencies: Eureka Server, Spring Web, Actuator. I have added the @EnableEurekaServer annotation on the main class like on the code below. With this annotation, the app will act like microservice registry and discovery server.

public class EurekaServerApplication {

    public static void main(String[] args) {, args);

The second Spring boot application will act as a client which will be registered on the eureka server (the first created application). This application needs to have the following dependencies: Actuator, Eureka Discovery, and Spring Web. We need to add the @EnableDiscoveryClient annotation on the Spring Boot app class and set up properties like on the code below into the application.yml file.

public class ClientDiscoveryApplication {

    public static void main(String[] args) {, args);

If we navigate on the following URL http://localhost:8761/ like on the image below. We can notice that 2 instances were registered on the eureka-client-service application and started on a different port. One of them was registered on port 8085 and the second instance in on port 8087.

Client-side discovery pattern

In this pattern, the client is responsible for determining the network locations of available service instances and load balancing requests. The network location of the service instance will be registered to the service registry when it start up and deregistered when it will be terminated.
The client is responsible for determining the network locations of available service instances and load balancing requests across them. The Service registry for client-side discovery which we are using in the previous example is Netflix Eureka Server. It provides management of service instances for querying available instances.

From the image below we can consider that we have 3 Services (Service A, Service B, and Service C ) which have different IP addresses provided to each service. For instance, the IP address of Service A is If we want to query the available instances, we need to access Service Registry (Eureka). The Eureka is responsible for storing the instances of all services that were previously registered.

  1. The locations of Service A, Service B and Service C are sent to the Service Registry like on the image below.
  2. The service consumer ask for the Service Discovery Server for the Location of Service A / Service B
  3. The location of Service A will be searched by the Service Registry which store the instance’s location
  4. The Service Customer gets the location of Service A and can make a direct request

Benefits of using this pattern
⦁ flexible, application specific load balancer
⦁ must implement register and discovery mechanism for each framework
The significant disadvantage of this pattern is that it couples the client with the service registry.

Alternatives to Eureka discovery

Other alternatives on Eureka are Consul and Zookeeper. Consul makes it simple for services to register themselves and to discover other services via DNS or HTTP interface. Also, provides a richer health checking. It has its own internal distributed key value store that can we use it as well.

Apache Zookeeper on the other side, is a distributed key value store. We can use it as the basis to implement service discovery. A centralized service for maintaining configuration information, naming, providing distributed synchronization. One of the benefit is that we can use it with a large community that supporting it.

Service-side discovery pattern

The alternative approach to Service Discovery is the Server-side Discovery pattern. Clients just make requests to a service instance via a load balancer, which acts as an orchestrator. The load balancer queries the service registry and routes each request to an available service instance like on the image below.
Benefits of using this pattern
⦁ no need to implement discovery logic for each framework used by service clients.
⦁ need to set up and manage the Load Balancer, unless it’s already provided in the deployment environment.
Some of the clustering solutions such as Kubernetes, run a proxy on each host in the cluster. In order to make a request to the service, a client connects to the local proxy using the port which is assigned to that service.

From the image below we can consider that we have added Load Balancer, that acts as an orchestrator. The Load Balancer queries the Service Registry and routes each HTTP request to an available service instance.


If we try to find some analogies between Kubernetes and more traditional architectures. I would compare Kubernetes Pods with service instances and Kubernetes Service with a logical set of pods.

Kubernetes Pods are the smallest, deployable unit in Kubernetes, which contain one or more containers, such as Docker containers. When a Pod runs multiple containers, the containers will be managed as a single entity and share the Pod’s resources. Each pod has its own IP address. When the pod restarts, it gets the new IP address. However there is no guarantee that the pod’s IP address will remain the same throughout all the time.

Kubernetes may relocate or re-instantiate pods at runtime. It doesn’t make sense to use the pod IP addresses for service discovery. This is one of the main reason to include Kubernetes Services into our story.

Service in Kubernetes

It is a component just like pod but it’s not a process. It’s just an abstraction layer which basically represents an IP address. Service will know to forward the request to one of the pods that have registered as service endpoints. On the other hand, unlike Kubernetes pods which don’t have the same IP address, the service has a stable IP address. Each service exposes an IP address and also exposes a DNS endpoint which will never change, even when the pod dies.

The Services provided by Kubernetes allow us to connect to our pods and provide a dynamic resource to access them. For instance, a load balancer can use these services to automatically determine which servers are trying to load balance. Services also provide load balancing ( if we have for example 3 instances of the microservice app, the service will get each request target to that and forward it to one of those pods).

In Kubernetes we have a concept of namespaces, which present a logical group of resources. For our examples I will using hello-today-dev namespace which consists of couples of pods.

kubectl -n hello-today-dev get pod -o wide

kubectl -n hello-today-dev get svc 

With this command we can see all available services in this namespace

How does Service Discovery work in Kubernetes?

Kubernetes has a powerful concept of labeling. Labels are just a key-value pair that provides metadata on our objects. Any pod whose labels match the selector defined in the service manifest will automatically be discover by a service.

Here we have a single service that is front-ending two of our pods. The two pods have a label named “app:my-app” and the Service has defined a label selector that is looking for those same labels.

If we have a single service that is front-ending two of our pods, instances of our app. These two pods have a label named “app=my-app” and the Service has defined a label selector that is looking for those same labels. This means that the service will send traffic to them, even thought the pods might change their addresses. You might also notice that there is a Pod3 that has a different label. The Service won’t front end that pod.

Example of service selector

The other example is when in a service selector we have defined two labels (app = my-app, microservice). Then service will looking for both of labels and it must match all the selectors, not just one and it will register as its endpoints. This is how the service will know which pods belong to it meaning where to forward that request to.

kubectl -n hello-today-dev get pods –show-labels

A service identifies its member pods using a selector attribute. We should specify a selector attribute which has a key-value pair defined as a list, which in our case is app = my-app. It creates a binding between the service and the pods with this name. It provides a flexible mechanism for service discovery, allowing us automatic load balancing. The Kubernetes will use the endpoints object to keep track of which pods are members of the service. Any pod whose label ‘s value (app=my-app) matches with the defined selector (my-app in our case) by the service will be exposed as its endpoint. Load Balancing will be provided by service by routing requests across matching pods.

kubectl -n hello-today-dev describe svc/ht-employee

The following command is to view the status of the service, in our case ht-employee. The service uses the selector app=ht-employee to select the pod as backend pods. The virtual IP address in the cluster is created for the Kubernetes Service to evenly distribute traffic to the pod at the backend.

We can see the defined endpoints of ht-employee microservice (that means that our service is working)

If we have black endpoints it is the result of not selecting any pods, which means that your service traffic will go nowhere. The Endpoints field indicates the pods specified by the selector field.


Depending of the deployment environments you needed, you can set up your own service discovery using service registry like Eureka. In other deployment environments as in our case is Kubernetes, service discovery is build in. The Kubernetes, as was explained previously is responsible for handling service instance registration and deregistration.

Deploy microservice on Kubernetes, step by step

Reading Time: 11 minutes

In this tutorial, I will try to explain step by step, how you can set up Kubernetes, deploy your microservice on Kubernetes, and check the result via the Kubernetes dashboard. All other things will be “as simple as possible”. As a cloud platform gcloud will be used. We will cover the following aspects of the problem:

  1. Create microservice to be deployed
  2. Placed application in your docker container
  3. What is Kubernetes and how to install it?
  4. Create a new Kubernetes project
  5. Create new Cluster
  6. Allow access from your local machine
  7. Create service account
  8. Activate service account
  9. Connect to cluster
  10. Gcloud initialization
  11. Generate access token
  12. Deploy and start Kubernetes dashboard
  13. Deploy microservice

Step 1: Create microservice to be deployed

Traditionally, in the programming world, everything starts with “Hello World”. So, as mentioned previously, to keep things simple, create a microservice that returns just “Hello World”. You can use for this goal. Create HelloController like this:

package com.example.demojooq.controllers;

import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;

public class HelloController {

    public String sayHello() {
        return "Hello world";

Step 2: Placed application in your docker container

We have a microservice, need to put this microservice in a docker container and upload it on Kubernetes. From that point, Kubernetes will orchestrate the container according to your settings. Let’s create the first image from the microservice. Normally, as you might guess, it is called Dockerfile (without any extension), and the content is:


FROM adoptopenjdk/openjdk11:jre-11.0.8_10-debianslim
ARG JAR_FILE=target/*.jar
COPY ${JAR_FILE} app.jar
ENTRYPOINT ["java","-jar","app.jar"]

The next step is to create the docker-compose file. For that purpose, a call to Dockerfile will be made to build the image. You can do it manually, but the best way is from the docker-compose file, as you have a permanent track of the solution. This is a .yaml file. (picture below)


version: "3"
    build: .
      - "8099:8080"

After starting docker, go to the folder where docker-compose is located and execute the command “docker-compose up”. The expectation is to reach this microservice on 8099 port. If everything is ok, in your docker will be something like this:

Check microservice docker installation with postman calling http://localhost:8099/api/v1/say-hello. In response, you have “Hello World”.

Step 3: What is Kubernetes and how to install it?

What is Kubernetes?

Kubernetes is an open-source container orchestrator that automates many tasks involved in deploying, managing, and scaling containerized applications. What happens when you use Docker, and your container fails? Probably the first thing to do is to restart your container. You do that manually (if you don’t have Kubernetes). Here comes Kubernetes, observe that container is down and start new container automatically. This is just a basic use case. Please read more on the internet, there is a bunch of information about this.

How to install Kubernetes?

Ok, until now you are sure that Kubernetes is needed, but where to find it, what are the costs, and how to install it? First of all, try “download Kubernetes” on Google… Pick the site… Options for Windows, Mac, Linux appear… A different installation like kind, minikube, kubeadmin… So, is it worth spending so much time setting properly this Kubernetes? You do not have to ask me, I agree with you, it is too much time. Fortunately, we can make a “go around” and skip all that: Go to Gcloud where Kubernetes is offered as a service and just use it. Somebody else takes care of this, we can focus just on the business logic in our microservice and use out-of-the-box Kubernetes installation from Gcloud. Sounds good, doesn’t it? The last and most important question; money. Is it for free? No, it is not. You have to pay for the Gcloud services and here is the price list: But for ordinary people like you and me, Gcloud offers a free account for 3 months up to 300$ and it seems fair. It is enough time to learn about deploying microservices on Kubernetes. For any professional use in future, the company should stay behind this. Here is the link where you can create your free cloud account One more thing, during the creation of a free account, Google will ask for your bank account, to automatically charge you. But do not worry, you are safe for the first three months and below 300$. And for any charging, you will be asked for permission before… So, until now my personal experience is positive, as Google is keeping the promise when you create an account. But the final decision is up to you.

Step 4: Create new Kubernetes project

Open up your Google account, sign in and go to the console.

Create a new project from the main dashboard; the name of the new project is “hello-world”. From now on, this is your active project.

Step 5: Create new cluster

Create new cluster (named it cluster2). Accept default values for others fields.

Step 6: Allow access from your local machine

Now, we must allow access from our local machine to Kubernetes, via kubectl. For that purpose, we need to follow these steps:

  1. Click on cluster2
  2. Find your local IP address and add it here according to the CIDR standard in the Edit control plane authorized networks

Step 7: Create service account

Give new account role “Owner”. Accept default values for other fields. After a service account is created, you should have something like this:

Generate keys for this service account with key type JSON. When the key is downloaded, it has some random name like hello-world-315318-ab0c74d58a70.json. Keep this file in a safe place, we will need it later.

Now, install Google Cloud SDK Shell on your machine according to your OS. Let’s do the configuration so kubectl can reach cluster2. Copy the file hello-world-315318-ab0c74d58a70.json and put it in the CLOUD SDK folder. For the Windows environment, it looks like this:

Step 8: Activate service account

The first thing to do is to activate the service account with the command: gcloud auth activate-service-account –key-file=hello-world-315318-ab0c74d58a70.json

Step 9: Connect to cluster

Now go to cluster2 again and find the connection string to connect to the new cluster

Execute this connection string in Google Cloud Shell: gcloud container clusters get-credentials cluster2 –zone us-central1-c –project hello-world-315318

Step 10: Gcloud initialization

The next command to execute is gcloud init, to initialize connection with the new project. Here is the complete code on how to do that from the Gcloud Shell:

C:\Users\Dimche Trifunov\AppData\Local\Google\Cloud SDK>gcloud init
Welcome! This command will take you through the configuration of gcloud.

Settings from your current configuration [dev] are:
  screen_reader: 'False'
  region: europe-west3
  zone: europe-west3-a
  disable_usage_reporting: 'True'
  project: dops-containers

Pick configuration to use:
 [1] Re-initialize this configuration [dev] with new settings
 [2] Create a new configuration
 [3] Switch to and re-initialize existing configuration: [database-connection]
 [4] Switch to and re-initialize existing configuration: [default]
Please enter your numeric choice:  2

Enter configuration name. Names start with a lower case letter and
contain only lower case letters a-z, digits 0-9, and hyphens '-':  hello-world
Your current configuration has been set to: [hello-world]

You can skip diagnostics next time by using the following flag:
  gcloud init --skip-diagnostics

Network diagnostic detects and fixes local network connection issues.
Checking network connection...done.
Reachability Check passed.
Network diagnostic passed (1/1 checks passed).

Choose the account you would like to use to perform operations for
this configuration:
 [7] Log in with a new account
Please enter your numeric choice:  5

You are logged in as: [].

API [] not enabled on project
[580325979968]. Would you like to enable and retry (this will take a
few minutes)? (y/N)?  y

Enabling service [] on project [580325979968]...
Operation "operations/acf.p2-580325979968-f1bf2515-deea-49d5-ae35-a0adfef9973e" finished successfully.
Pick cloud project to use:
 [1] hello-world-315318
 [2] Create a new project
Please enter numeric choice or text value (must exactly match list
item):  1

Your current project has been set to: [hello-world-315318].

Do you want to configure a default Compute Region and Zone? (Y/n)?  n

Error creating a default .boto configuration file. Please run [gsutil config -n] if you would like to create this file.
Your Google Cloud SDK is configured and ready to use!

* Commands that require authentication will use by default
* Commands will reference project `hello-world-315318` by default
Run `gcloud help config` to learn how to change individual settings

This gcloud configuration is called [hello-world]. You can create additional configurations if you work with multiple accounts and/or projects.
Run `gcloud topic configurations` to learn more.

Some things to try next:

* Run `gcloud --help` to see the Cloud Platform services you can interact with. And run `gcloud help COMMAND` to get help on any gcloud command.
* Run `gcloud topic --help` to learn about advanced features of the SDK like arg files and output formatting

Step 11: Generate access token

Type kubectl get namespace, access token is generated in .kube folder (in home folder), in config file:

If you open this config file, you will find your access token. You will need this later.

Step 12: Deploy and start Kubernetes dashboard

Now, deploy Kubernetes dashboard with the next command: kubectl apply -f

C:\Users\Dimche Trifunov\AppData\Local\Google\Cloud SDK>kubectl apply -f
namespace/kubernetes-dashboard created
serviceaccount/kubernetes-dashboard created
service/kubernetes-dashboard created
secret/kubernetes-dashboard-certs created
secret/kubernetes-dashboard-csrf created
secret/kubernetes-dashboard-key-holder created
configmap/kubernetes-dashboard-settings created created created created created
deployment.apps/kubernetes-dashboard created
service/dashboard-metrics-scraper created
deployment.apps/dashboard-metrics-scraper created

C:\Users\Dimche Trifunov\AppData\Local\Google\Cloud SDK>kubectl proxy
Starting to serve on

Start the dashboard with kubectl proxy command. Now open the dashboard from the link: http://localhost:8001/api/v1/namespaces/kubernetes-dashboard/services/https:kubernetes-dashboard:/proxy/#/overview?namespace=default

In front of you, this screen will appear:

Now, you need the token from the config file that we spoke about a moment ago. Open the config file with Notepad (on Windows), find your access token, and copy from there and paste it in the Enter token* field. Be careful when you are copying token from the config file as there might be several tokens. You must choose yours (image below).

Finally, the stage is prepared to deploy microservice.

Step 13: Deploy microservice

Build the docker image from Dockerfile with the command: docker build -t docker2222/dimac:latest. docker2222/dimac is my public docker repository.
Push the image on docker hub with the command: docker image push docker2222/dimac:latest.
Execute kubectl apply -f k8s.yaml where k8s.yaml is the file below:


apiVersion: v1
kind: Namespace
  name: hello


apiVersion: apps/v1
kind: Deployment
  name: hello-world
  namespace: hello
    buildNumber: "1.0"
      app: hello-world
  replicas: 1
        app: hello-world
        buildNumber: "1.0"
        - name: hello-world
          image: docker2222/dimac:latest
              path: "/actuator/health/readiness"
              port: 8080
            initialDelaySeconds: 5
            - containerPort: 8080
            - name: APPLICATION_VERSION
              value: "1.0"

apiVersion: v1
kind: Service
  name: hello-world
  namespace: hello
    app: hello-world
    - protocol: TCP
      port: 80
      targetPort: 8080

Last but not least, open the Kubernetes dashboard. If everything is OK, you should see your service.

How we deploy with Terraform and BitBucket to Azure Kubernetes

Reading Time: 6 minutes

N47 implemented a set of back-office web applications for Prestige, a real estate management company located in Zurich, Switzerland. One application is a tool for displaying construction projects nearby properties managed by Prestige and a second example is a tool for creating and assigning orders to craftsmen. But the following examples aren’t specific for those use cases.

Screenshot of the Construction Project tool.

An Overview

The project entails one frontend application with multiple microservices whereby each service has its own database schema.

The application consumes data from Prestige’s main ERP system Abacus and third-party applications.

N47 is responsible for setting up and maintaining the full Kubernetes stack, MySQL Database, Azure Application Gateway and Azure Active Directory applications.

Another company is responsible for the networking and the Abacus part.

Architectural Overview

Involved Technologies

Our application uses following technologies:

  • Database: MySQL 8
  • Microservices: Java 11, Spring Boot 2.3, Flyway for database schema updates
  • Frontend: Vue.js 2.5 and Vuetify 2.3
  • API Gateway: ngix

The CI/CD technology stack includes:

  • Source code: BitBucket (Git)
  • Pipelines: BitBucket Pipelines
  • Static code analysis: SonarCloud
  • Infrastructure: Terraform
  • Cloud provider: Azure

We’ll focus on the second list of technologies.

Infrastructure as Code (IaC) with Terraform and BitBucket Pipelines

One thing I really like when using IaC is having the definition of the involved services and resources of the whole project in source code. That enables us to track the changes over time in the Git log and of course, it makes it far easier to set up a stage and deploy safely to production.

Please read more about Terraform in our blog post Build your own Cloud Infrastructure using Terraform. The Terraform website is of course as well a good resource.

Storage of Terraform State

One important thing when dealing with Terraform is storing the state in an appropriate place. We’ve chosen to create an Azure Storage Account and use Azure Blob Storage like this:

terraform {
  backend azurerm {
    storage_account_name = "prestigetoolsterraform"
    container_name = "prestige-tools-dev-tfstate"
    key = "prestige-tools-dev"

The required access_key is passed as an argument to terraform within the pipeline (more details later). You can find more details in the official tutorial Store Terraform state in Azure Storage by Microsoft.

Another important point is not to run pipelines in parallel, as this could result in conflicts with locks.

Used Terraform Resources

We provide the needed resources on Azure via BitBucket + Terraform. Selection of important resources:

Structure of Terraform Project

We created an entry point for each stage (local, dev, test and prod) which is relatively small and mainly aggregate to the modules with some environment-specific configurations.

The configurations, credentials and other data are stored as variables in the BitBucket pipelines.


The modules themselves have always a file, and to have a clean separation of input, logic and output.

Example source code of the azure_aplication_insights module (please note that some of the text have been shortened in order to have enough space to display it properly)

variable "name" {
  type = string

variable "location" {
  type = string

variable "resource_group_name" {
  type = string

resource "azurerm_application_insights" "ai" {
  name                =
  location            = var.location
  resource_group_name = var.resource_group_name
  application_type    = "web"

output "instrumentation_key" {
  value =

BitBucket Pipeline

The BitBucket pipeline controls Terraform and includes the init, plan and apply. We decided to manually apply the changes in the infrastructure in the beginning.

image: hashicorp/terraform:0.12.26

      - step:
        name: Plan DEV
          - cd environments/dev
          - terraform init -backend-config="access_key=$DEV_TF_CONFIG_ACCESS_KEY"
          - terraform plan -out out-overall.plan
          - environments/dev/out-overall.plan

      - step:
        name: Plan DEV
          - cd environments/dev
          - terraform init -backend-config="access_key=$DEV_TF_CONFIG_ACCESS_KEY"
          - terraform plan -out out-overall.plan
          - environments/dev/out-overall.plan
          - environments/dev/.terraform/**
      - step:
        name: Apply DEV
        trigger: manual
        deployment: dev
          - cd environments/dev
          - terraform apply out-overall.plan

      - step:
          name: Plan TEST
            - cd environments/test
            - terraform init -backend-config="access_key=$PRESTIGE_TF_CONFIG_ACCESS_KEY"
            - terraform plan -out out-overall.plan
            - environments/test/out-overall.plan
            - environments/test/.terraform/**
      - step:
          name: Apply TEST
          trigger: manual
          deployment: test
            - cd environments/test
            - terraform apply out-overall.plan

      # PRESTIGE PROD ...

Needed Steps for Deploying to Production

1. Create feature branch with some changes

2. Push to Git (BitBucket pipeline with step Plan DEV will run). All the details about the changes can be found in the Terraform plan command

3. Create a pull request and merge the feature branch into develop. This will start another pipeline with the two steps (plan + apply)

4. Check the output of the plan step before triggering the deploy on dev

5. Now the dev stage is updated and if everything is working as you wish, create another pull request to merge from develop to master. And re-do the same for the production of other stages

We have just deployed an infrastructure change to production without logging into any system except BitBucket. Time for celebration.

people watching concert
Symbol picture of N47 production deployment party (from unsplash)

Is Really Everything That Shiny?

Well, everything is a big word.

We found issues, for example with cross-module dependencies, which aren’t just solvable with a depends_on. Luckily, there are some alternatives:

network module:

output "id" {
  description = "The Azure assigned ID generated after the Virtual Network resource is created and available."
  value =

kubernetes cluster module, which depends on network:

variable "subnet_depends_on" {
  description = "Variable to force module to wait for the Virtual Network creation to finish"

and the usage of those two modules in environments/dev/

module "network" {
  source = "../../modules/azure_network"

module "kubernetes_cluster" {
  source = "../../modules/azure_kubernetes_cluster"
  subnet_depends_on =

After having things set up, it really makes joy to wipe out a stage and just provision everything from scratch with running a BitBucket pipeline.