Flowable 6 adds ad-hoc sub process support

With the first official release of Flowable 6 approaching we have added support for ad-hoc sub processes. It's a feature that shows the strength of the Engine refactoring we have been doing, because it's a fairly simple implementation on an Engine level. Let's have a look at what's possible with ad-hoc sub processes in Flowable 6.

We haven't updated the BPMN editor yet with the ad-hoc sub process symbols and properties, as this is an Engine feature in its current state. But the editor part will be added as well of course. So let's imagine the following embedded sub process to be ad-hoc.

The ad-hoc sub process contains two tasks without any sequence flow being defined. When a process instance is started from this process definition, an ad-hoc sub process execution is created in the Flowable Engine and no user task is created yet. For an ad-hoc sub process you can get a list of enabled activities, which refers to the activities that are available within the ad-hoc sub process to work on. So you can work on Task 1 and Task 2 in any order. With the following API call you can get a list of enabled activities from the Engine.

runtimeService.getEnabledActivitiesFromAdhocSubProcess(executionId);

This gives back a list of FlowNode objects that are enabled in the ad-hoc sub process for which the execution identifier is passed as parameter. In this example there will be two UserTask objects returned by this API call. So you can imagine a UI / application where a user can choose which task to work on next. The API call to start one of these tasks looks like this:

runtimeService.executeActivityInAdhocSubProcess(executionId, id);

So again the execution identifier of the ad-hoc sub process is provided, and in addition an identifier of the user task to select is provided as well. This identifier matches the id attribute of the user task element in the BPMN XML of the process definition. When this call is executed for Task 1 for example, the Flowable Engine will create a new user task, which will be available just like any other user task. 

An ad-hoc sub process by default can execute multiple enabled activities at the same time. This is defined with the ordering attribute in the BPMN XML definition. This means that we could also execute Task 2 in our example, while also having Task 1 active at the same time.

And what happens when Task 1 and/or Task 2 are completed. Without defining a completionCondition attribute in the BPMN XML, the Flowable Engine will not end the ad-hoc sub process execution automatically. The RuntimeService API provides the following method to complete an ad-hoc sub process when there are no active child executions (for example user tasks) anymore:

runtimeService.completeAdhocSubProcess(executionId);

After invoking this method our example process will continue to the After task activity and the ad-hoc sub process will be ended.

But invoking a method to complete an ad-hoc sub process isn't appropriate for a lot of use cases. So let's look at an BPMN XML example with a defined completion condition.

<adHocSubProcess id="adhocSubProcess" ordering="Sequential">

    

    <userTask id="subProcessTask" name="Task in subprocess" />

    <userTask id="subProcessTask2" name="Task2 in subprocess" />

      

    <completionCondition>${completed}</completionCondition>

    

</adHocSubProcess>

There are two differences with the previous example process definition. The first difference is the ordering attribute, which is set to Sequential. This means that only one of the user tasks can be executed at the same time. The Flowable Engine will not allow a second user task to be executed, when the first user task hasn't been completed yet. The second difference is the completion condition. This ad-hoc sub process will be automatically completed when the ${completed} expression evaluates to true. So, when we execute the first user task in the ad-hoc sub process and complete it together with a variable completed of true, the ad-hoc will end automatically after completing the user task.

Until now we have been demonstrating individual user tasks within an ad-hoc sub process only. But it's also possible to define individual task together with flows as well. Let's look at an example process model.

In this example Task 1 and Task 2 are enabled when the ad-hoc sub process is started by the Flowable Engine. So it's not possible to execute the other user tasks yet. When Task 1 is executed and completed, the Next task will be automatically created by the Engine, just like it would happen in a normal flow in a process definition. Using the completion condition, you can still determine when the ad-hoc sub process can be completed. And using the cancelRemainingInstances attribute it's possible to define whether the ad-hoc sub process should cancel any remaining executions / tasks when the completion condition evaluates to true. By default the Flowable Engine will cancel all other running executions / tasks, but when setting this attribute to false, the ad-hoc sub process will not complete before all executions / tasks have been ended.

The Flowable 6 Engine opens up a whole new range of dynamic behaviour in process instances. Because of its well defined execution structure and flexible API, there are a lot of new possibilities to add to the Engine. Ad-hoc sub processes is just an example as you will see more functionality appearing in Flowable 6 in the near future.

You can already play around with the ad-hoc sub process functionality on the master branch on Github (https://github.com/flowable/flowable-engine). We are looking for feedback from the community about this feature and other BPMN related features you would like to see implemented in Flowable 6.

Adding dynamic behaviour to the Flowable Engine

We have added a new feature to the Flowable Engine that I would like to share with you even before it's available as part of an official release. The new feature is the ability to change a defined number of properties of a process definition without needing to redeploy the process definition.

Before I'll go into details about this feature, some examples of needing to redeploy a process definition would be good to set the context. One example is a service task with a defined class name in the process definition. This service task is part of a big process definition, and process instances can be active for over a year. After deploying the first version of the process definition to the production environment the requirements of the process execution have changed and a new Java class should be executed for the service task instead. This means that new deployment of the process definition is needed and existing process instances need to be migrated to this new process definition.

Another example is a user task with a candidate group management. Over time, the requirements have changed and the sales group needs to be set as the candidate group for this user task. Again, this would mean a new deployment of the process definition and migration of existing process instances.

With the addition of the DynamicBpmnService interface these use cases can be solved differently, without the need to redeploy the process definition. For example, the DynamicBpmnService interface has a method to change the class name of a specific service task named changeServiceTaskClassName. Let's look at a code example to change the class name of the exampleTask service task.

<serviceTask id="exampleTask" 
  flowable:class="org.flowable.PreviousTask" />

ObjectNode infoNode = dynamicBpmnService.changeServiceTaskClassName(
      "exampleTask", "org.flowable.NewTask");
dynamicBpmnService.saveProcessDefinitionInfo("procDefId", infoNode);

After the invocation of the saveProcessDefinitionInfo method the Flowable Engine will now execute the NewTask class instead of the PreviousTask defined in the BPMN XML. As you can see we are using Jackson JSON for storing the overriding process definition information in the Flowable database. Every process definition can have zero or one entry in the new ACT_PROCDEF_INFO table. All changes for a process definition are defined in the same JSON object and stored as a BLOB in the Flowable database. The cache that's used to retrieve the process definition info data is versioned and designed to work in a clustered Flowable environment.

Let's look at one more example, changing the candidateGroup of a user task:

<userTask id="exampleTask" flowable:candidateGroup="management" />

ObjectNode infoNode = dynamicBpmnService.changeUserTaskCandidateGroup(
      "exampleTask", "sales", true);
dynamicBpmnService.saveProcessDefinitionInfo("procDefId", infoNode);

Again, after saving the JSON object, the Flowable Engine will use the sales candidate group for the exampleTask user task instead of the management group.

If you want to override multiple properties of a specific process definition in different steps you need to get the currently stored JSON object first. This is easy to do like this:

ObjectNode infoNode = dynamicBpmnService.getProcessDefinitionInfo("procDefId");

These are the properties we currently support to change with this approach:

1. Service task - class name
2. Service task - expression
3. Service task - delegate expression
4. User task - name
5. User task - description
6. User task - due date
7. User task - priority
8. User task - category
9. User task - form key
10. User task - assignee
11. User task - owner
12. User task - candidate users
13. User task - candidate groups

And there's an additional feature we have implemented with the same approach, which is localisation. It's now possible to store the user task name and description in multiple languages in the process definition info table and when you query for tasks you can set a locale to use. The returned tasks will then contain the localised name and description as defined in the process definition info table. Let's look at an example:

<userTask id="exampleTask" name="Name" />

ObjectNode infoNode = dynamicBpmnService.changeLocalizationName(
      "es", "exampleTask", "Nombre");
dynamicBpmnService.saveProcessDefinitionInfo("procDefId", infoNode);

Task task = taskService.createTaskQuery().locale("es").singleResult();

The returned task instance will now have Nombre as the name of the task.

We are really excited about this new feature that's available in both the Flowable 5 and 6 code that's in Github, and we are eager to get feedback about this new feature. Let us know what you think about it and maybe come up with suggestions of which properties we could add to this approach in addition to the current supported list. We'll do a 5.18.1 release in a couple weeks time, together with a new beta release for Flowable 6, that will include this feature. And you can already create your own build using the code on Github of course.

Mule 3 released

It maybe a bit strange to start of this blog with a post about Mule, but the release of version 3.0 of the Mule community edition is certainly a valid reason.

It's been a while since I wrote the book Open-source ESBs in Action with Jos Dirksen. But the coverage of Mule in that book was based on Mule 2.x, which was the current version until a few weeks. In the last couple of years other open source integration frameworks like Apache Camel and Spring Integration came along and provided similar functionality. With version 3.0 Mule steps up again and provides great additional functionality with foremost the following features: hot deployment, new flow based architecture, annotation support and better support for web services and rest.

I want to highlight two features of Mule 3.0 in this post, which are the new flow based architecture and hot deployment. Previous versions of Mule had a service based architecture to implement your integration logic. The following picture taken from the Mule userguide describes it in a very easy manner.

As the picture shows, integration logic was implemented via an inbound router to process incoming messages, via a service component to implement additional processing logic, and an outbound router for sending the message along the path. In XML this looked liked the following snippet:

<service name="hello">
  <inbound>
    <jms:inbound-endpoint queue="hello.in"/>
  </inbound>
  <component class="org.mule.TestComponent"/>
  <outbound>
    <pass-through-router>
      <jms:outbound-endpoint queue="hello.out"/>
    </pass-through-router>
  </outbound>
</service>

Nothing wrong with this kind of architecture, but it becomes a bit tricky when you have to specify all kinds of transfers, multiple component classes etc. The new flow based architecture solves these issues and provides a very clean and flexible architecture based on message processors and sources to implement your integration logic (see the following figure taken from the userguide).

All the things you want to do with a message after it has arrived at a message source is implemented with message processors. So transformers, routers and custom logic are all message processors. This makes it very clean and easier to understand. The new flow based architecture looks like this in XML:

<flow name="hello">
  <jms:inbound-endpoint queue="hello.in"/>
  <component class="org.mule.TestComponent"/>
  <jms:outbound-endpoint queue="hello.out"/>
</flow>

This looks very simple doesn't it? For a very small example like I've implemented here, the differences are not huge of course. But you can imagine that for more complex examples, the new flow based architecture makes your life a lot easier.

The other huge improvement in Mule 3.0 is hot deployment. One of the main differences between Apache ServiceMix and Mule when we wrote the Open-source ESBs in Action book was a hot deployment feature from an enterprise perspective. And now with version 3.0 it's there and looks very promissing. In the apps directory of the Mule installation you can now deploy a Mule configuration and JARs (if necessary) and Mule picks it up automatically. Then, if you want to change the Mule configuration you can simply update the file and again Mule picks it up automatically. For more information you can look at the userguide. Hot deployment now works for a lot of implementations, but there is still some room for improvement. For example, only the main Mule configuration file is monitored by Mule, not its child Mule configurations. When compared to other open source ESBs like Apache ServiceMix and Open ESB, I see that Mule takes a web application like approach to hot deployment, where the others use OSGi.

So to summarize, version 3.0 is certainly a step forward! It provides a lot of improvements and some great new features including the discussed hot deployment and the new flow based architecture. And from a perspective of the open source BPM Activiti project I'm enthousiastic about the possibilities of integration between Mule and Activiti. So I hope to see Activiti integration in the next 3.x version of Mule.