EventBridge relaunches custom buses with ordering, retention and CloudEvents
Victoria Shinder2 min readAmazon Web Services announced on 24 September a new version of the EventBridge custom event bus, the serverless broker that routes events between applications, SaaS integrations and AWS services. The enhanced bus adds several capabilities that event-driven systems previously had to assemble from other services.
A custom bus can now be shared with other accounts through AWS Resource Access Manager, so publishers send events directly to a central bus. A new publishing API accepts events in common JSON formats such as CloudEvents and preserves their schema without modification. Events are retained for 24 hours by default, extendable to one year, so consumers can recover from errors or new components can be hydrated from history. A new Subscriber resource filters events and delivers them to more than 250 AWS services. The bus now supports strictly ordered processing, with events handled in the exact order received, and content-based deduplication.
The existing bus is renamed "Custom event bus - classic", and AWS says all existing APIs remain unchanged. The enhanced bus launches in fourteen regions across the United States, Europe and Asia Pacific, with a new pricing model that charges for data transferred rather than per event, with lower costs as workloads scale.
What it means
Retention, replay and ordering are features teams usually reach for Kafka or Kinesis to get. Adding them to EventBridge narrows the gap between a routing service and a streaming log, and for organisations that already use EventBridge as the glue between teams it may remove the need to run a second system just for history or ordering.
The pricing change deserves attention before migrating. Charging by data rather than by event favours many small events and penalises large payloads, so the bill for an existing workload can move in either direction. The classic bus staying unchanged means there is no forced migration; the new bus is a choice to make per workload.