Most of the companies nowadays are implementing or want to implement architectures based on micro-services. While this can help companies to overcome multiple challenges, it can bring its own new challenges to the table.

In this article, we are going to discuss a very concrete one been maintaining compatibility when we change objects that help us to communicate the different micro-services on our systems. Sometimes, they are called API objects, Data Transfer Objects (DTO) or similar. And, more concretely, we are going to be using Jackson and JSON as a serialisation (marshalling and unmarshalling) mechanism.

There are some other methods, other technologies and other ways to achieve this but, this is just one tool to keep on our belt and be aware of to make informed decisions when faced with this challenge in the future.

Maintaining compatibility is a very broad term, to establish what we are talking about and ensure we are on the same page, let us see a few examples of real situations we are trying to mitigate:

• To deploy breaking changes when releasing new features or services due to changes on the objects used to communicate the different services. Especially, if at deployment time, there is a small period of time where the old and the new versions are still running (almost impossible to avoid unless you stop both services, deploy them and restart them again).
• To be forced to have a strict order of deployment for our different services. We should be able to deploy in any order and whenever it best suits the business and the different teams involved.
• The need of, due to a change in one object in one concrete service, being forced to deploy multiple services not directly involved or affected by the change.
• Related to the previous point, to be forced to change other services because of a small data or structural change. An example of this would be some objects that travel through different systems been, for example, enriched with extra information and finally shown to the user on the last one.

To exemplify the kind of situation we can find ourselves in, let us take a look at the image below. In this scenario, we have four different services:

• Service A: It stores some basic user information such as the first name and last name of a user.
• Service B: It enriches the user information with extra information about the job position of the user.
• Service C: It adds some extra administrative information such as the number of complaints open against the user.
• Service D: It finally uses all the information about the user to, for example, calculate some advice based on performance and area of work.

All of this is deployed and working on our production environment using the first version of our User object.

At some point, product managers decided the age field should be considered on the calculations to be able to offer users extra advice based on proximity of retirement. This added requirement is going to create a second version of our User object where the field age is present.

Just a last comment, for simplicity purposes, let us say the communication between services is asynchronous based on queues.

As we can see on the image, in this situation only services A and D should be modified and deployed. This is what we are trying to achieve and what I mean by maintaining compatibility. But, first, let us explore what are the options we have at this point:

1. Upgrade all services to the second version of the object User before we start sending messages.
2. Avoid sending the User from service A to service D, send just an id, and perform a call from service D to recover the User information based on the id.
3. Keep the unknown fields on an object even, if the service processing the message at this point does not know anything about them.
4. Fail the message, and store it for re-processing until we perform the upgrade to all services involved. This option is not valid on synchronous communications.

Option 1

As we have described, it implies the update of the dependency service-a-user in all the projects. This is possible but it brings quickly some problems to the table:

• We not only need to update direct dependencies but indirect dependencies too what it can be hard to track, and easy to miss. In addition, a decision needs to be done about what to do when a dependency is missed, should an error be thrown? Should we fail silently?
• We have a problem with scenarios where we need to roll back a deployment due to something going wrong. Should we roll back everything? Good luck! Should we try to fix the problem while our system is not behaving properly?
• Heavy refactoring operations or modifications can make upgrades very hard to perform.

Option 2

Instead of sending the object information on the message, we just send an id to be able posteriorly to recover the object information using a REST call. This option while very useful in multiple cases is not exempt from problems:

• What if, instead of just a couple of enrichers, we have a dozen of them and they need the user information? Should we consolidate all services and send ids for the enriched information crating stores on the enrichers?
• If, instead of a queue, other mechanisms of communications are used such as RPC, do now all the services need to call service A to recover the User information and do their job? This just creates a cascade of calls.
• And, under this scenario, we can have inconsistent data if there is any update while the different services are recovering a User.

Option 3

This is going to be the desired option and the one we are going to do a deep dive on this article using Jackson and JSON how to keep the fields even if the processing service does not know everything about them.

To add in advance that, as always, there are no silver bullets, there are problems that not even this solution can solve but it will mitigate most of the ones we have named on previous lines.

One problem we are not able to solve with this approach – especially if your company perform “all at once” releases instead of independent ones – is, if service B, once deployed tries to persist some information on service A before the new version has been deployed, or tries to perform a search using one criterion, in this case, the field age, on the service A. In this scenario, the only thing we can do is to throw an error.

Option 4

This option, especially in asynchronous situations where messages can be stored to be retried later, can be a possible solution to propagate the upgrade. It will slow down our processing capabilities temporarily, and retrying mechanism needs to be in place but, it is doable.

Using Jackson to solve versioning

Renaming a field

Plain and simple, do not do it. Especially, if it is a client-facing API and not an internal one. It will save you a lot of trouble and headaches. Unfortunately, if we are persisting JSON on our databases, this will require some migrations.

If it needs to be done, think about it again. Really, rethink it. If after rethinking it, it needs to be done a few steps need to be taken:

1. Update the API object with the new field name using @JsonAlias.
2. Release and update everything using the renamed field, and @JsonAlias for the old field name.
3. Remove @JsonAlias for the old field name. This is a cleanup step, everything should work after step two.

Removing a field

Like in the previous case, do not do it, or think very hard about it before you do it. Again, if you finally must, a few steps need to be followed.

First, consider to deprecate the field:

If it must be removed:

1. Explicitly ignore the old property with @JsonIgnoreProperties.
2. Remove @JsonIgnoreProperties for the old field name.

Unknown fields (adding a field)

Ignoring them

The first option is the simplest one, we do not care for new fields, a rare situation but it can happen. We should just ignore them:

A note of caution in this scenario is that we need tone completely sure we want to ignore all properties. As an example, we can miss on APIs that return errors as HTTP 200 OK, and map the errors on the response if we are not aware of that, while in other circumstances it will just crash making us aware.

Ignoring enums

In a similar way, we can ignore fields, we can ignore enums, or more appropriately, we can map them to an UNKNOWN value.

Keeping them

The most common situation is that we want to keep the fields even if they do not mean anything for the service it is currently processing the object because they will be needed up or downs the stream.

Jackson offers us two interesting annotations:

• @JsonAnySetter
• @JsonAnyGetter

These two annotations help us to read and write fields even if I do not know what they are.

class User {
    @JsonAnySetter
    private final Map<String, Object> unknownFields = new LinkedHashMap<>();
    
    private Long id;
    private String firstname;
    private String lastname;

    @JsonAnyGetter
    public Map<String, Object> getUnknownFields() {
        return unknownFields;
    }
}

Keeping enums

In a similar way, we are keeping the fields, we can keep the enums. The best way to achieve that is to map them as strings but leave the getters and setters as the enums.

@JsonAutoDetect(
    fieldVisibility = Visibility.ANY,
    getterVisibility = Visibility.NONE,
    setterVisibility = Visibility.NONE)
class Process {
    private Long id;
    private String state;

    public void setState(State state) {
        this.state = nameOrNull(state);
    }

    public State getState() {
        return nameOrDefault(State.class, state, State.UNKNOWN);
    }

    public String getStateRaw() {
        return state;
    }
}

enum State {
    READY,
    IN_PROGRESS,
    COMPLETED,
    UNKNOWN
}

Worth pointing that the annotation @JsonAutoDetect tells Jackson to ignore the getters and setter and perform the serialisation based on the properties defined.

Unknown types

One of the things Jackson can manage is polymorphism but this implies we need to deal sometimes with unknown types. We have a few options for this:

Error when unknown type

We prepare Jackson to read an deal with known types but it will throw an error when an unknown type is given, been this the default behaviour:

@JsonTypeInfo(
    use = JsonTypeInfo.Id.NAME,
    include = JsonTypeInfo.As.PROPERTY)
@JsonSubTypes({
    @JsonSubTypes.Type(value = SelectionProcess.class, name = "SELECTION_PROCESS"),
})
interface Process {
}

Keeping the new type

In a very similar to what we have done for fields, Jackson allow as to define a default or fallback type when the given type is not found, what put together with out unknown fields previous implementation can solve our problem.

@JsonTypeInfo(
    use = JsonTypeInfo.Id.NAME,
    include = JsonTypeInfo.As.PROPERTY,
    property = "@type",
    defaultImpl = AnyProcess.class)
@JsonSubTypes({
    @JsonSubTypes.Type(value = SelectionProcess.class, name = "SELECTION_PROCESS"),
    @JsonSubTypes.Type(value = SelectionProcess.class, name = "VALIDATION_PROCESS"),
})
interface Process {
    String getType();
}

class AnyProcess implements Process {
    @JsonAnysetter
    private final Map<String, Object> unknownFields = new LinkedHashMap<>();

    @JsonProperty("@type")
    private String type;

    @Override
    public String getType() {
        return type;
    }

    @JsonAnyGetter
    public Map<String, Object> getUnknownFields() {
        return unknownFields
    }
}

And, with all of this, we have decent compatibility implemented, all provided by the Jackson serialisation.

We can go one step further and implement some basic classes with the common code e.g., unknownFields, and make our API objects extend for simplicity, to avoid boilerplate code and use some good practices. Something similar to:

class Compatibility {
    ....
}

class MyApiObject extends Compatibility {
    ...
}

With this, we have a new tool under our belt we can consider and use whenever is necessary.