This is a Java, Maven, Akka project that demonstrates how to setup an Akka Cluster with an example implementation of Cluster Sharding running in a Kubernetes cluster.
This project is one in a series of projects that starts with a simple Akka Cluster project and progressively builds up to examples of event sourcing and command query responsibility segregation.
The project series is composed of the following projects:
- akka-java-cluster
- akka-java-cluster-aware
- akka-java-cluster-singleton
- akka-java-cluster-sharding
- akka-java-cluster-persistence
- akka-java-cluster-persistence-query
- akka-java-cluster-kubernetes (this project)
Each project can be cloned, built, and runs independently of the other projects.
This project contains an example implementation of cluster sharding running in a Kubernetes cluster. The project also includes a live web client visualization of the cluster and the Akka actors running in the cluster JVMs.
Once this application is successfully deployed to a properly configured Kubernetes cluster you will be able to pull up a live view of the running system, as shown above in Figure 1.
The above visualization is rendered using the D3Js library. The web page used to render the visualization is located in src/main/resources/monitor.html.
The monitor webpage is accessed via the HTTP server that is running in the HttpServerActor class. The HTTP server is started in this actor so in addition to handling HTTP requests this actor also handles incoming actor messages.
The monitor web page uses a web socket to communicate with the HTTP server. The primary action is that every few seconds, the web page sends a request for the current status of the running cluster. The HTTP server actor responds with a JSON tree that contains the current cluster state.
This tree starts with a single root element that represents the cluster. The next tree elements up from the root represent Akka cluster nodes. These elements also represent actors that are started one per cluster node. An example of a single actor per cluster node is the HttpServerActor. One instance of this actor is started on each Akka cluster node.
In the visualization shown in Figure 1, the three large circles represent the Akka cluster nodes. These circles are shown using the colors brown, orange, and yellow. A yellow circle indicates which node is receiving HTTP requests from the browser. A brown circle is used to show where cluster singleton actors are currently located in the cluster. Orange circles show other cluster nodes. You can mouseover each large circle to see what each one represents.
Continuing up from the Akka node tree elements are shard actor elements. Shard actors are used in Akka Cluster Sharding to distribute entity actor instances across the cluster nodes. In the visualized tree the shard actors are shown as green circles. There are a fixed number of shard actors. The number of shard actors is defined in the application.conf file. In the visualization, you will see that the fixed number of shard actors will redistribute themselves across the cluster and the number of cluster nodes changes.
The tree leaf elements represent individual entity actors. The entity actors are shown as blue circles. When an entity actor instance is first started, the color used is a darker blue. When an entity actor instance is stopped the color fades to pink and then the circle disappears.
There are several options for how we set up a running Kubernetes environment. You can install everything on your local development system, or you can use one fo the cloud-based offerings. In addition to setting up a Kubernetes environment, you will also need to install Docker on your local development system. Of course, you will also need to install Java 8 and Maven 3.6.x.
"There are various ways to get free updates (including security), and (new and existing) paid support models available from various vendors to consider." - Source Java is Still Free.
You can find Java downloads at the followinf locations:
The ApaptOpenJDK is recommended as this is the JDK that Lightbend engineerin teams build against.
Maven 3.6.x is available for download on the Apache Maven download page.
This project includes two Maven plugins. One of the plugins creates a self-contained JAR file that contains all of the compiled project class files and all of the dependencies. The other plugin creates a Docker image that contains the JAR file and the necessary settings that are used to run the Java code.
Follow the link for the Docker installation for your device.
Instructions and download are available on the Install MiniKube page.
Instructions and download are available on the Welcome to MiniShift page.
Go to the OpenShift Online site at https://manage.openshift.com.
Create an account or login to your existing Red Hat account.
Select an OpenShift Online plan. For testing you have the option to select the Free Plan. The Free Plan gets you access to the 60-day sandbox.
Follow the directions to setup an OpenShift environment. Once the environment is ready open the Web Console.
On the top right of the OpenShift Online Console click the arrow to the right of your user name. In the dropdown menu click "Copy Login Command." This will open a new browser tab. Click the "Display Token" link. Copy the "oc login ..." command and paste it into a command line window. Now the OpenShift CLI is connected to the OpenShift environment.
# cd into the project directory
cd akka-java-cluster-kubernetes
# Build the project, which creates a self contained JAR file and then loads it into a Docker image
mvn clean package docker:build
# Create the akka-cluster-1 project
oc new-project akka-cluster-1 --display-name="Akka Cluster Demo" --description="A demonstration of Akka cluster sharding application running on Kubernetes"
# Select the project as the currnet project
oc project akka-cluster-1
# Login to the OpenShift containter registry
oc registry login
# Login Docker into the container registry
docker login $(oc registry info) -u $(oc whoami) -p $(oc whoami -t)
# Tag the Docker image in preparation for uploading into the container registry
docker tag akka-cluster-demo:1.0.0 $(oc registry info)/akka-cluster-1/akka-cluster-demo:1.0.0
# Push the tagged Docker image into the container registry
# This creates an image stream named akka-cluster-demo
docker push $(oc registry info)/akka-cluster-1/akka-cluster-demo:1.0.0
# Create the rolebinding used for Akka cluster bootstrap
oc apply -f kubernetes/akka-cluster-openshift-online-rolebindging.yml
# Create the network service and route to the HTTP endpoint on port 8080
oc apply -f kubernetes/akka-cluster-openshift-online-service-route-http.yml
# Use thw following to create the image name to be used in the deployment YAML file
# Edit the akka-cluster-deployment.yml file changing the image: on line 30 with the image name from the below echo command
echo $(oc registry info --internal=true)/$(oc project -q)/akka-cluster-demo:1.0.0
# Deploy the Akka cluster demo app
oc apply -f kubernetes/akka-cluster-deployment.yml
# If all is well, you should see 3 running or starting pods
oc get pods
# This is the URL of the application web page
echo http://$(oc get route akka-cluster-demo -o=jsonpath="{.status.ingress[0].host}")Use the OpenShift console to review the project and the running app.
On the console, select the Administrator user on the top left. Then select Networking and click Routes. On the Routes view on the right click the URL under Location. This shouls open a tab that will show the app visualization.
Go to IBM Cloud at https://cloud.ibm.com/.
Create an account or login to an existing IBM account. Select a plan. Free plans are available.
From the service Catalog at https://cloud.ibm.com/catalog, select Kubernetes Service.
Follow the directions for Creating Kubernetes clusters.
# cd into the project directory
cd akka-java-cluster-kubernetes
# Build the project, which creates a self contained JAR file and then loads it into a Docker image
mvn clean package docker:build
# Create the akka-cluster-1 project
oc new-project akka-cluster-1 --description="Akka Java Cluster Kubernetes Example" --display-name="akka-cluster-1"
# Select the project as the currnet project
oc project akka-cluster-1
# Tag the Docker image in preparation for uploading into the container registry
docker tag akka-cluster-demo:1.0.0 <your-ibm-cloud-repo-name>/akka-cluster-demo:1.0.0
# Push the tagged Docker image into the container registry
docker push <your-ibm-cloud-repo-name>/akka-cluster-demo:1.0.0
# Create the rolebinding used for Akka cluster bootstrap
oc apply -f kubernetes/akka-cluster-rolebindging.yml
# Deploy the Akka cluster demo app
oc apply -f kubernetes/akka-cluster-deployment.yml
# Make the HTTP endpoint port externally accessible
oc expose deployment/akka-cluster-demo --type=NodePort --port 8080
# Show the HTTP endpoint info created avoe
oc get services/akka-cluster-demo- Make sure all environment variables are configured properly. You should be able to run the following commands:
docker --versionoc version(openShift installation) orkubectl version(Kubernetes only installation) Note: both command require a running environment.minishift --version(orminikube --version) and that you have access to all the executables which will be necessary, namely:docker,kubectl,minishift(orminikube)
- Start the docker service if it isn't running already (if using Docker Toolbox this should require running "Docker Quickstart Terminal" (with administrator privileges if applicable))
- If using
minishift, before running theminishift startcommand for the first time, make sure you have set the desired vm-driver (if applicable), with a command such as:minishift config set vm-driver virtualbox(for virtualbox driver)
- The first run may take a while because it will:
- Download OpenShift
ocbinary - Download MiniShift ISO and start MiniShift VM
- Pull OpenShift Docker Container Image and start OpenShift Cluster
- Download OpenShift
- Run
minishift start - Take note of the final output of that command, which should be something similar with the following:
The server is accessible via web console at:
https://192.168.99.100:8443/console
You are logged in as:
User: developer
Password: <any value>
To login as administrator:
oc login -u system:admin- Verify that you can access the web console and login as developer in the web console, and as system:admin in your terminal/shell window
- If using
minishift, obtain the commands you need to type into your shell in order to add theocbinary to yourPATHenvironment variable:
minishift docker-envminishift oc-env- Then run the respective commands. For example:
eval $(minishift docker-env)
eval $(minishift oc-env)- Make sure you login from your terminal/shell with:
oc login -u system:admin - The kubernetes YAML files used throughout this project, refer to the
akka-cluster-1namespace. First, we need to create it, in one of two ways:
- Using OpenShift/MiniShift:
- Login as
developerin the Web UI and create a new project manually, with the appropriate name to match the intended namespace - Or create it from the console make sure you are loggin as
developer:- If you followed the steps until this point, you should be
- If you previously ran:
oc login -u system:admin, you can logout that user with:oc logout, and then, if necessary, runoc login, specifyingdeveloperas the user, and<anything>as the password, and then:oc new-project akka-cluster-1 --description="Akka Java Cluster Kubernetes Example" --display-name="akka-cluster-1"- This will create a project in OpenShift/MiniShift, which will include the respective namespace and will allow you to see its related content in the web UI
- Login as
- Alternatively, if not using OpenShift/MiniKube, or if you just don't want the UI integration, you can create the namespace with
kubectl create namespace akka-cluster-1
- Login as admin in the terminal/shell with:
oc login -u system:admin - Specify the namespace in use for the subsequent commands in this example by running:
- For OpenShift/MiniShift -
oc project akka-cluster-1 - For Minikube or plain
kubectlusage -kubectl config set-context --current --namespace=akka-cluster-1(alternatively, most kubectl commands can take a--namespace akka-cluster-1flag to explicitly reference this namespace)
NOTE: In some cases when running download-intensive steps during workshops where multiple attendees are using the same network, sometimes the traffic from the external IP may be blocked. When this is the case, it is recommended to use a VPN connection in order to be able to proceed.
The following command builds the project, loads all of the code and dependencies into a single JAR file, and then creates a Docker image that is set up to run the JAR file in a Docker container.
When running Kubernetes on your laptop, the instructions for each specific set up define how Docker is used when Docker images are created.
mvn clean package docker:buildThe Docker image is deployed to a Kubernetes environment using the following two steps.
First, a Role and RoleBinging are set up. The Role and RoleBinding give the pods access to the API server. More details are provided in the Kubernetes API documentation, which is part of Akka Cluster Bootstrap. The Role and RoleBinding only need to be created once.
kubectl apply -f kubernetes/akka-cluster-rolebindging.ymlOr, for example, deploy using MiniShift.
oc apply -f kubernetes/akka-cluster-rolebindging.ymlThe second step is to deploy the Docker image.
oc apply -f kubernetes/akka-cluster-deployment.ymlThree Kubernetes pods are started once the Docker image is deployed using the deployment YAML file.
$ kubectl get pods
NAME READY STATUS RESTARTS AGE
akka-cluster-demo-56c6c46cb4-5zvpl 1/1 Running 16 22d
akka-cluster-demo-56c6c46cb4-khl6g 1/1 Running 6 14d
akka-cluster-demo-56c6c46cb4-thscr 1/1 Running 8 22dAfter the Kubernetes pods are started the next step is to expose the visualization web page to an accessible network location.
oc expose deployment/akka-cluster-demo --type=NodePort --port 8080The visualization web page should now be accessible. Use the following commands to obtain the web page port and the IP address.
export NODE_IP=$(minishift ip) # or: $(minikube ip)
export NODE_PORT=$(kubectl get services/akka-cluster-demo -o go-template='{{(index .spec.ports 0).nodePort}}')
echo http://$NODE_IP:$NODE_PORTUse the above URL to access the visualization web page.
Once the cluster is running, you can simulate the elasticity and resilience of the Akka cluster running in a Kubernetes environment. With Kubernetes, you can configure a cluster to auto scale the number of running pods based on CPU and memory utilization. With the cluster visualization, you can demonstrate what happens when the cluster is scaled up or down by manually scaling up or down the number of running pods.
kubectl scale --replicas=5 deployment/akka-cluster-demoUse the kubectl or oc command to view the number of running pods.
$ kubectl get pods
NAME READY STATUS RESTARTS AGE
akka-cluster-demo-56c6c46cb4-5hk7p 1/1 Running 0 1m
akka-cluster-demo-56c6c46cb4-5zvpl 1/1 Running 16 22d
akka-cluster-demo-56c6c46cb4-khl6g 1/1 Running 6 14d
akka-cluster-demo-56c6c46cb4-lx9mv 1/1 Running 0 1m
akka-cluster-demo-56c6c46cb4-thscr 1/1 Running 8 22dWith auto-scaling, the number of running pods are adjusted as the CPU or memory loads change. Manually you can scale down the number of nodes to simulate this and then use the cluster visualization to view what is happening both at the Kubernetes level and the Akka level.
$ kubectl scale --replicas=1 deployment/akka-cluster-demo
deployment.extensions/akka-cluster-demo scaled
$ kubectl get pods
NAME READY STATUS RESTARTS AGE
akka-cluster-demo-56c6c46cb4-5hk7p 1/1 Terminating 0 8m
akka-cluster-demo-56c6c46cb4-5zvpl 1/1 Terminating 16 22d
akka-cluster-demo-56c6c46cb4-khl6g 1/1 Running 6 14d
akka-cluster-demo-56c6c46cb4-lx9mv 1/1 Terminating 0 8m
akka-cluster-demo-56c6c46cb4-thscr 1/1 Terminating 8 22dA short time later the terminating pods should be gone.
$ kubectl get pods
NAME READY STATUS RESTARTS AGE
akka-cluster-demo-56c6c46cb4-khl6g 1/1 Running 6 14dClick one of the pod circles in the visualization to simulate the loss of a pod. Clicking a pod will trigger the JVM to stop, which will trigger Kubernetes to restart the pod. Visually this is shown where the clicked pod and all of the associated actors disappear. Kubernetes reacts by restarting the pod. In the visualization, you will see a new pod appear after a brief period.
The visualization of the cluster presents an interesting problem. In the visualization, actors are shown from across the cluster. However, browser requests from the visualization web page of course land on a single cluster node. So the challenge is how to show all of the activity from across the cluster from a single node. The solution is using cluster-aware actors. A cluster-aware actor knows, by design, that there are instances of itself on each node in the cluster. When one of the cluster-aware actors receives a message, this can trigger that actor to send messages to the other instances of itself. In the case of the visualization, when an EntityActor starts or stops, it sends a message to the HttpServerActor entity. These entity messages are used to add or remove the corresponding tree elements that are used to show entities in the visualization tree.
private void notifyStart() {
EntityMessage.Action start = new EntityMessage.Action(memberId, shardId, entityId, "start", true);
httpServer.tell(start, self());
}
private void notifyStop() {
EntityMessage.Action stop = new EntityMessage.Action(memberId, shardId, entityId, "stop", true);
httpServer.tell(stop, self());
}The EntityActor contains two methods, notifyStart() and notifyStop(). These methods are invoked when an instance of an entity actor instance is started or stopped. Note that this actor sends a message to the HttpServerActor.
private void actionEntity(EntityMessage.Action action) {
log().info("Entity {} <-- {}", action, sender());
if (action.action.equals("start")) {
tree.add(action.member, action.shardId, action.entityId);
} else if (action.action.equals("stop")) {
tree.remove(action.member, action.shardId, action.entityId);
}
if (action.forward) {
forwardAction(action.asNoForward());
}
}When an HttpServerActor receives a start or stop message from an entity actor, it invokes the actionEntity(...) method. Note that the tree object adds or removes the entity based on the information provided in the actor message. Then, if the message forward flag is true, the forwardAction(...) method is invoked.
private void forwardAction(Object action) {
cluster.state().getMembers().forEach(member -> {
if (!cluster.selfMember().equals(member) && member.status().equals(MemberStatus.up())) {
forwardAction(action, member);
}
});
}
private void forwardAction(Object action, Member member) {
String httpServerPath = member.address().toString() + self().path().toStringWithoutAddress();
ActorSelection httpServer = context().actorSelection(httpServerPath);
log().debug("{} --> {}", action, httpServer);
httpServer.tell(action, self());
}There are two overloaded forwardAction(...) methods. The first method loops through all of the nodes in the cluster. In the loop, cluster members are filtered to select only node members that are in the up state and are not this node.
The second forwardAction method builds an actor selection, which is similar to actor references, using the passed member info and the path part of this actor's actor reference. Note that the forward flag in the forwarded message is set to false.
So the flow here is that as each EntityActor starts or stops it sends an action message to an HttpServerActor, which happens to be an instance of the HttpServerActor running on the same node as the EntityActor. The HttpServerActor is cluster aware, it has the necessary code that forwards the action messages on to the other HttpServerActors running on each node in the cluster.
# Note, in the commands below, "kubectl" can be used instead of "oc" as an alternative or if you are using minikube
# See the details of the exposed service
$ oc get services
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
akka-cluster-demo NodePort 172.30.106.6 <none> 8080:32531/TCP 9m
# Remove it
oc delete service akka-cluster-demo
# Delete the deployment
oc delete -f kubernetes/akka-cluster-deployment.yml
# Delete the rolebinding
oc delete -f kubernetes/akka-cluster-rolebindging.yml
# Delete the project/namespace
## If using OpenShift/MiniShift
oc delete project akka-cluster-1
## If using Minikube
kubectl delete namespace akka-cluster-1You can also stop the container services previously started with:
minishift stop # or: minikube stop
docker-machine stop default # assuming that it is the "default" container running