Skip to main content
Version: 8.5

Handling data in processes

When using Camunda, you have access to a dynamic map of process variables, which lets you associate data to every single process instance (and local scopes in case of user tasks or parallel flows). Ensure you use these mechanisms in a lightweight and meaningful manner, storing just the relevant data in the process instance.

Depending on your programming language, consider accessing your process variables in a type safe way, centralizing (simple and complex) type conversion and using constants for process variable names.

Understanding data handling in Camunda

When reading and interpreting a business process diagram, you quickly realize there is always data necessary for tasks, but also to drive the process through gateways to the correct next steps.

Examine the following tweet approval process example:


The process instance starts with a freshly written tweet we need to remember.


We need to present this tweet so that the user can decide whether to approve it.


The gateway needs to have access to this information: was the tweet approved?


To publish the tweet, the service task again needs the tweet itself!

Therefore, the tweet approval process needs two variables:

Variable nameVariable typeSample value
tweetString"@Camunda rocks"

In Camunda 8, values are stored as JSON.

Camunda 7 handles variables slightly differently

This best practice describes variable handling within Camunda 8. Process variables are handled slightly differently with Camunda 7. Consult the Camunda 7 documentation for details. In essence, variable values are not handled as JSON and thus there are different values supported.

You can dynamically create such variables by assigning an object of choice to a (string typed) variable name; for example, by passing a Map<String, Object> when completing the "Review tweet" task via the API:

// TODO: Double check!
taskId: "547811"
variables: [
name: "approved"
value: true

In Camunda, you do not declare process variables in the process model. This allows for a lot of flexibility. Refer to recommendations below on how to overcome possible disadvantages of this approach.

Consult the docs about variables to learn more.

Camunda does not treat BPMN data objects () as process variables. We recommend using them occasionally for documentation, but you need to avoid excessive usage of data objects.

Storing just the relevant data

Do not excessively use process variables. As a rule of thumb, store as few variables as possible within Camunda.

Please note the technical limitations of variables sizes.

Storing references only

If you have leading systems already storing the business relevant data...

Hold references only

...then we suggest you store references only (e.g. ID's) to the objects stored there. So instead of holding the tweet and the approved variable, the process variables would now, for example, look more like the following:

Variable nameVariable typeValue

Use cases for storing payload

Store payload (actual business data) as process variables, if you....

  • ...have data only of interest within the process itself (e.g. for gateway decisions).

In case of the tweet approval process, even if you are using a tweet domain object, it might still be meaningful to hold the approved value explicitly as a process variable, because it serves the purpose to guide the gateway decision in the process. It might not be true if you want to keep track in the tweet domain objects regarding the approval.

Variable nameVariable typeValue
  • ...communicate in a message oriented style. For example, retrieving data from one system and handing it over to another system via a process.

When receiving external messages, consider storing just those parts of the payload relevant for you, and not the whole response. This not only serves the goal of having a lean process variables map, it also makes you more independent of changes in the service's message interface.

  • ...want to use the process engine as kind of cache. For example, you cannot query relevant customer data in every step for performance reasons.

  • ...need to postpone data changes in the leading system to a later step in the process. For example, you only want to insert the Tweet in the Tweet Management Application if it is approved.

  • ...want to track the historical development of the data going through your process.

  • ...don't have a leading system for this data.

Using constants and data accessors

Avoid the copy/paste of string representations of your process variable names across your code base. Collect the variable names for a process definition in constants. For example, in Java:

public interface TwitterDemoProcessConstants {
String VAR_NAME_TWEET = "tweet";
String VAR_NAME_APPROVED = "approved";

This way, you have much more security against typos and can easily make use of refactoring mechanisms offered by your IDE.

However, if you also want to solve necessary type conversions (casting) or probably even complex serialization logic, we recommend that you use a Data Accessor class. It comes in two flavors:

  • A Process Data Accessor: Knows the names and types of all process variables of a certain process definition. It serves as the central point to declare variables for that process.
  • A Process Variable Accessor: Encapsulates the access to exactly one variable. This is useful if you reuse certain variables in different processes.

Consider, for example, the BPMN "Publish on Twitter" task in the Tweet Approval Process:


We use a TweetPublicationDelegate to implement the "Publish on Twitter" task:

public class PublishTweetJobHandler implements JobHandler  {
public void handle(JobClient client, ActivatedJob job) throws Exception {
String tweet = job.getVariablesAsType(TwitterDemoProcessVariables.class).getTweet();
// ...

The tweet variable is accessed in a type safe way.

This reusable Process Data Accessor class could, for example, be a simple object. The Java client API can automatically deserialize the process variables as JSON into this object, while all process variables that are not found in that class are ignored.

public class TwitterDemoProcessVariables {

private String tweet;
private boolean approved;

public String getTweet() {
return tweet;

public void setTweet(String tweet) {
this.tweet = tweet;

The getters and setters could further take care of additional serialization and deserialization logic for complex objects.

Your specific implementation approach might differ depending on the programming language and framework you are using.

Complex data as entities

There are some use cases when it is clever to introduce entities alongside the process to store complex data in a relational database. You can observe this logically as typed process context where you create custom tables for your custom process deployment. Then, you can even use Data Accessor classes to access these entities in a convenient way.

You will only store a reference to the entity's primary key (typically an artificial UUID) as real process variable within Camunda.

Some people refer to this as externalized process context.

There are a couple of advantages of this approach:

  • You can do very rich queries on structured process variables via typical SQL.
  • You can apply custom data migration strategies when deploying new versions of your process or services, which require data changes.
  • Data can be designed and modeled properly, even graphically by, for example, leveraging UML.

It requires additional complexity by adding the need for a relational database and code to handle this.