As distributed systems evolve, the need may arise to change which endpoint handles a specific message type. During the development phase, such changes are relatively straightforward. But when the system is in production, additional considerations are necessary for messages that are in-flight (i.e. which have been sent but not yet consumed).
How to move handlers between endpoints
To move a handler from
SourceEndpoint to a
DestinationEndpoint, take the following steps:
- Copy the handler into
DestinationEndpointand deploy that endpoint. Now both endpoints are handling the same message type.
- Make sure that
DestinationEndpointreceives the messages that it should be handling. From now on all newly sent messages should only go to the DestinationEndpoint.
- After ensuring that the
SourceEndpointdoes not have any remaining messages of the type that should be processed by
DestinationEndpoint, delete the handler from
SourceEndpointand deploy it.
How to ensure that the messages get to the DestinationEndpoint
This depends on whether the message is a command or an event.
When the handler processes events, it is important for the
DestinationEndpoint to subscribe to that event. That will ensure that the event is delivered to
DestinationEndpoint. After that,
SourceEndpoint should unsubscribe from the event to ensure that no new events are delivered to the
When the handler processes commands, every piece of code that sends the command must be updated to send it to a queue of
DestinationEndpoint rather than
SourceEndpoint. In the meantime, the
SourceEndpoint handler may be changed to forward the messages to the new destination.
How to handle error messages that were discovered after removing the handler
When a set of messages are found that were sent to a
SourceEndpoint and a handler was already removed, a retry redirect can be set up to redirect every failed message to a queue for
DestinationEndpoint. After retrieving those messages, the redirect retry should be removed.
Reasons to move handlers between endpoints
An endpoint that handles multiple message types may not be able to keep up with the volume of messages that it receives. One way to mitigate this is to separate high throughput handlers into separate endpoints, i.e. to divide the load into more endpoints.
When an endpoint processes multiple message types, some of the messages might need to be processed faster than others; that is, service level agreements for a given message type could differ from others. In those cases, a simple solution is to move the processing of that message to a separate endpoint so that those message types don't have to compete with other message types in the queue.