Limits of ROS2 Message Schema Evolution: Unresolved Rollback Decision

The Short Answer

ROS2 cannot currently guarantee a safe rollback of message schemas without redeploying nodes because it relies on Common Data Representation (CDR) for serialization. CDR uses fixed field offsets; any change to the message structure (except adding fields to the end of a message) shifts these offsets, causing binary incompatibility. Since the schema is baked into the node at compile-time, a node expecting a new schema will fail to deserialize data from a rolled-back node providing an old schema.

Technical Explanation: CDR and Compatibility

The core of the issue lies in how ROS2 leverages DDS. While some DDS implementations support flexible schema evolution, the ROS2 rosidl generator produces static C++/Python structures. This creates two distinct failure modes during rollbacks:

• Backward Compatibility: A new node can often read an old message if fields were only added to the end, as the deserializer simply finds the data stream ended prematurely and treats missing fields as default values.
• Forward Compatibility (The Rollback Trap): When you roll back a publisher to an older version, a newer subscriber still expects the extended data stream. This often results in "unexpected end of data" errors or, worse, silent data corruption if the field types align but the semantic meaning has changed.

Confirmed Constraints

Change Type	Compatibility	Rollback Risk
Adding fields to end	Partial (Backward)	High (Subscriber expects data)
Changing field type	None	Critical (Deserialization failure)
Reordering fields	None	Critical (Offset misalignment)

Recommended Mitigation Steps

To achieve a pseudo-rollback capability without full redeployment, avoid modifying existing .msg files. Instead, follow these patterns:

1. Versioned Topics: Instead of evolving /sensor_data, publish to /sensor_data_v2. This allows you to keep both versions of a node running and switch between them via parameters.
2. Opaque Blobs: For highly volatile data, use a string or sequence<uint8> field to wrap a flexible serialization format like Protobuf or JSON, handling the versioning logic within the application layer.
3. Side-by-Side Deployment: Deploy the new node version as a separate executable and use a lifecycle manager to transition traffic from the old version to the new one.

Verification Method

To verify compatibility before a rollout, use two separate colcon workspaces:

# Workspace A: Old Schema
# Workspace B: New Schema
# Run Publisher (A) → Subscriber (B) [Tests Backward Compatibility]
# Run Publisher (B) → Subscriber (A) [Tests Forward Compatibility/Rollback]

Use ros2 topic echo /topic_name to observe if the middleware reports deserialization errors or truncated messages.

Diagnostic Detail Needed: Are you using a specific DDS vendor (e.g., FastDDS, CycloneDDS) that has been configured with custom TypeSupport, or are you relying on the standard ROS2 rosidl generated types?