Optimize your problems using KIE Execution Server

Fri 18 August 2017

Matej Čimbora

Ex OptaPlanner Workbench developer

KIE Server provides separation of execution environment from your client application, making it easy to use OptaPlanner outside Java ecosystem (e.g. .NET). In the 7 release stream, we worked to improve OptaPlanner/KIE Server integration. The new API is less verbose and more intuitive. Let’s take a closer look at its usage in more detail.

How a typical workflow looks like?

Install your optimization app kjar to a maven repository accessible by the KIE Server. This step assumes the following conditions are met:
- kmodule.xml file is located in META-INF folder of your project.
- Project uses packaging type kjar.
- kie-maven-plugin is configured for the project.
```
  <build>
    <plugins>
      <plugin>
        <groupId>org.kie</groupId>
        <artifactId>kie-maven-plugin</artifactId>
        <version>${version.org.kie}</version>
        <extensions>true</extensions>
      </plugin>
    </plugins>
  </build>
```

Create a container (server unit of execution) wrapping the project. This example uses Java KIE Server client, which performs REST calls in the background.

The following dependencies are required in your client application. Make sure to use version 7.0.0.Final, or later.

  <dependency>
    <groupId>org.kie.server</groupId>
    <artifactId>kie-server-api</artifactId>
  </dependency>
  <dependency>
    <groupId>org.kie.server</groupId>
    <artifactId>kie-server-client</artifactId>
  </dependency>

Client and container creation:

  KieServicesConfiguration kieServicesConfiguration = new KieServicesConfigurationImpl(SERVER_URL, USERNAME, PASSWORD, CLIENT_TIMEOUT);
  KieServicesClient kieServicesClient = KieServicesFactory.newKieServicesClient(kieServicesConfiguration)

  // ReleaseId corresponds to groupId:artifactId:version (GAV) of the project installed in the first step
  ReleaseId releaseId = new ReleaseId("org.optaplanner", "kie-server-example", "1.0.0-SNAPSHOT");
  KieContainerResource containerResource = new KieContainerResource(CONTAINER_ID, releaseId);
  kieServicesClient.createContainer(CONTAINER_ID, containerResource);

Create a solver within the container. A single container can hold multiple solvers, each of them having a unique identifier. This is useful if you want to try out different solver configurations.
```
SolverServicesClient solverClient = kieServicesClient.getServicesClient(SolverServicesClient.class);
solverClient.createSolver(CONTAINER_ID, SOLVER_ID, SOLVER_CONFIG_XML_PATH);
```

Submit your optimization problem

CloudBalance cloudBalance = loadPlanningSolution();
solverClient.solvePlanningProblem(CONTAINER_ID, SOLVER_ID, cloudBalance);

Check the solver state. At this point of time the best solution notifications are not supported yet on the client side, you need to periodically check the solver status/best solution.

  SolverInstance solver = solverClient.getSolver(CONTAINER_ID, SOLVER_ID);
  if (solver.getStatus() == SolverInstance.SolverStatus.SOLVING) {
    // continue
  } else {
    CloudBalance cloudBalance = (CloudBalance) solverClient.getSolverWithBestSolution(CONTAINER_ID, SOLVER_ID).getBestSolution()
    // process the solution
  }

Once you are done with the optimization, terminate the solver.

  solverClient.terminateSolverEarly(CONTAINER_ID, SOLVER_ID);

If you don’t plan to reuse the solver anymore, dispose it to free the resources on the server.
```
  solverClient.disposeSolver(CONTAINER_ID, SOLVER_ID);
```

New OptaPlanner REST API overview

The following table shows an overview of the new API introduced in version 7.0.0.Final. All solver action URLs are normally prefixed by container URL, e.g. http://${kie-server}/services/rest/server/containers/${container_id}, which is not included in the table to keep the output short.

Action	Method	URL suffix	Payload
Register solver	PUT	solvers/${solver_id}	-
Get solver	GET	solvers/${solver_id}	SolverInstance object
Submit solution	POST	solvers/${solver_id}/state/solving	Solution object
Get best solution	GET	solvers/${solver_id}/bestsolution	SolverInstance object including the best solution
Terminate solver	POST	solvers/${solver_id}/state/terminating-early	-
Dispose solver	DELETE	solvers/${solver_id}	-

Action

Method

URL suffix

Payload

PUT

solvers/${solver_id}

Get solver

GET

solvers/${solver_id}

SolverInstance object

Submit solution

POST

solvers/${solver_id}/state/solving

Solution object

Get best solution

GET

solvers/${solver_id}/bestsolution

SolverInstance object including the best solution

Terminate solver

POST

solvers/${solver_id}/state/terminating-early

Dispose solver

DELETE

solvers/${solver_id}

Where do I start?

Download KIE (Execution) Server.

Check KIE Server Integration chapter of OptaPlanner Workbench quickstart or OptaPlanner REST API documentation for more details on the API.

Optionally try out KIE Workbench, which integrates with the KIE Server. The following video demonstrates the process of setting up a Workbench example and optimizing it using the KIE Server.

Conclusion

KIE Server provides an alternative way of optimizing your constraint satisfaction problems. If you want to offload workload to cloud (possibly with a large amount of resources), or you use non-Java client, KIE Server is the right pick.

Comments Permalink