Skip to content

Latest commit

 

History

History
209 lines (148 loc) · 11.9 KB

design.md

File metadata and controls

209 lines (148 loc) · 11.9 KB

Dekorate Design

Overview

This section provides a high level overview on the design of Dekorate. The core building blocks are back-quoted and a definition for them is provided in the next section (vocabulary).

Dekorate provides multiple annotation processors all targeting at the generation of Kubernetes resources. Each annotation processor is "supporting" one or more annotations and will be called by the compiler at least twice when the annotations are spotted.

On each annotation processor invocation, information about the annotated classes are passed. The last invocation signals the end of the processing.

When the processing is over, each processor will assemble a config object and pass it to the responsible handler. The handlers role is to create or modify the model based on the config.

Finally, the model is serialized to disk as json/yml.

[sequence diagram]

Vocabulary

The section below describes the core parts of Dekorate and tries to describe how they work together.

Visitor

Refers to the the Gang of Four visitor pattern. As Kubernetes resources are deeply nested we are extensively using the visitor pattern to perform modifications to those resources without having to programmatically traverse this complex structure.

Model

With Kubernetes/OpenShift model or just model we refer to the java object representation of the Kubernetes/OpenShift resource domain. The model has the same structure are the actual kubernetes resources and can be easily serialized into json or yml and form the actual resources.

The base interface that all model objects implement is the HasMetadata. For example:

 public class Pod implemnets HasMetadata {
 
   private String kind;
   private ObjectMeta metadata;
   
   public String getKind() {
     return this.kind;
   }
    
   public ObjectMeta getMetadata() {
      return this.metadata;
   }
 }

Config

An object / pojo that encapsulates the information provided by an annotation and the project (e.g. name, version etc). An example configuration is the KubernetesConfig class:

public class KubernetesConfig  {
    private String group;
    private String name;
    private String version;
    private io.dekorate.kubernetes.config.Label[] labels;
    private io.dekorate.kubernetes.config.Annotation[] annotations;
    ...
}

ConfigurationGenerator

A service that given properties extracted from sources like:

  • annotations
  • local configurations
  • framework

will convert them into a Config using a mapper and by applying various Configurators (see below).

Configurator

A configurator is a visitor that visits parts of the config with the purpose of performing minor changes / updates.

For example a configurator that can be used to add a label to KubernetesConfig:

public class AddLabel extends Configurator<KubernetesConfigFluent> {

     public void visit(KubernetesConfigFluent config) {
         config.addToLabels(new Label("createdBy", "dekorate");
     }
}
Configurator Target Description
ApplyOpenshiftConfig SourceToImageConfig Applies group, name and version from OpenshiftConfig to SourceToImage config.
AddPort KubernetesConfig Adds a port to all containers.
ApplyDockerBuildHook DockerBuildConfig Apply the docker build hook configuration.
ApplySourceToImageHook SourceToImageConfig Apply source to image build hook.

ManifestGenerator

An object that processes a certain type of config with the purpose of generating a manifest. In some cases a ManifestGenerator may not generate the whole manifest, but influence parts of it. Usually such generators create model resources and register decorators.

 public class KubernetesManifestGenerator extends AbstractKubernetesManifestGenerator<KubernetesConfig> implements ManifestGenerator {

    public void generate(KubernetesConfig config) {
        resourceRegistry.add("kubernetes", createDeployment(config));
    }

    public boolean accepts(Class<? extends Configuration> type) {
      return type.equals(KubernetesConfig.class);
    }
}

Decorator

A decorator is a visitor that visits parts of the kubernetes/openshift model in order to perform minor changes / updates. It's different than a configurator in the sense that it operates on the actual model instead of the config.

public class AddLabel extends Decorator<PodFluent> {

     public void vist(PodFluent podFluent) {
         podFluent.addToLabels(new Label("createdBy", "dekorate");
     }
}

The decorator looks pretty similar to the configurator the only difference between the two being the kind of objects they visit.

Configurators visit config objects. Decorators visit model objects.

Decorator Target Description
AddSecretVolume PodSpec Add a secret volume to all pod specs.
AddService KubernetesList Add a service to the list.
AddEnvVar Container Add a environment variable to the container.
AddLivenessProbe Container Add a liveness probe to all containers.
AddReadinessProbe Container Add a readiness probe to all containers.
AddStartupProbe Container Add a startup probe to all containers.
AddConfigMapVolume PodSpec Add a configmap volume to the pod spec.
AddEnvToComponent ComponentSpec Add environment variable to component.
AddAzureDiskVolume PodSpec Add an Azure disk volume to the pod spec.
AddAnnotation ObjectMeta A decorator that adds an annotation to all resources.
AddMount Container Add mount to all containers.
AddPort Container Add port to all containers.
AddPvcVolume PodSpec Add a persistent volume claim volume to all pod specs.
AddAwsElasticBlockStoreVolume PodSpec Add an elastic block store volume to the pod spec.
AddRuntimeToComponent ComponentSpec Add the runtime information to the component.
AddLabel ObjectMeta Add a label to the all metadata.
AddAzureFileVolume PodSpec Add an Azure File volume to the Pod spec.

Why do we need both Configurators and Decorators?

There are two main reasons:

  • decoupling configuration sources from manifest generation logic
  • working on a simplified representation of the model
Decoupling configuration sources from manifest generator logic

There are multiple configuration sources that contribute to the actual configuration that will be used for manifest generation:

  • Annotations
  • Property configuration
  • Framework artifacts
  • Defaults

Instead of coupling all those sources with the manifest generation logic, it is preferable to limit their scope to the creation of the config object. So pratctically we are having two main processing phases:

  • configuration creation
  • manifest generation

Since, the config objects themselves are fluent buildable objects, that support the visitor pattern it felt natural to use it. To distinguish between visitors that are acting on configuration vs visitors that are acting on the generated manifests we have Configurator and Decorator classes respectively.

Using a simplified model representation where it makes sense

The kubernetes model is very complex and deeply nested object structure and for a good reason: It needs to fit to every signle deployment use case out there. The deployment of a java application though is something more concrete and can be described by something simpler than the actual model.

During the process of gathering and combining information from multiple annotation processors its more practical and less error prone to apply them to a more simplified representation of the model, which is what the config essentially is. So, during the processing phase we use configurators to apply the information gathered in each step to the config.

Once the configuration is finalized, the actual model is populated. Since different processors are creating different kinds of config we need to combine them all in order to build the model. This is where decorators come in place. Each config is translated to different decorators that contribute to different parts of the model.

Why not directly creating the model?

We have a variable number of config instances all contributing to the model. Combining them all in one go without the use of decorators would result in a conditional hell.

Annotation Processor

Refers to Java annotation processors. Each processor is responsible for creating a config object and also for registering one or more handler that handle the config. A processor may register more than one config handlers with no restriction on the kind of config they handle.

Processor Config Supported Annotations Description
KubernetesAnnotationProcessor KubernetesConfig [io.dekorate.kubernetes.annotation.KubernetesApplication] Generates kubernetes manifests.
OpenshiftAnnotationProcessor OpenshiftConfig [io.dekorate.kubernetes.annotation.KubernetesApplication, io.dekorate.openshift.annotation.OpenshiftApplication] Generates openshift manifests.
SpringBootApplicationProcessor none [org.springframework.boot.autoconfigure.SpringBootApplication] Detects Spring Boot and set the runtime attribute to Spring Boot.
ThorntailProcessor none [javax.ws.rs.ApplicationPath, javax.jws.WebService] Detects JAX-RS and JAX-WS annotations and registers the http port.
SpringBootMappingProcessor none [org.springframework.web.bind.annotation.RequestMapping, org.springframework.web.bind.annotation.GetMapping] Detects Spring Boot web endpoints and registers the http port.

Session

A shared repository between annotation processors. This repository holds ConfigurationGenerators and ManifestGenerators. When the session is closed, All ConfigurationGenerators will be used to populate the config that is passed to the ManifestGenerators. The ManifestGenrators generate and decorate the model. The resulting model is passed back to the processor.