Migrating from Amazon MQ to self-run RabbitMQ without downtime
Move queues, exchanges and consumers off Amazon MQ with the RabbitMQ project's own blue-green method, into a cluster in our cloud, your cloud account or your data center.
- ✓Operated by AceMQ
- ✓160+ customers
To migrate from Amazon MQ to RabbitMQ without downtime, build a new RabbitMQ cluster, import the definitions, link the two brokers with federation or shovel, move consumers, then producers, and drain the old queues. The RabbitMQ project documents this as a blue-green deployment. Amazon MQ's plugin rules decide which side defines the link.
Which migration method fits
RabbitMQ documents a blue-green deployment for moving to a new cluster: import the topology, link the clusters so messages flow to the new side, move consumers and producers, then retire the old cluster. Federation and shovel are the two link types it uses, and they keep both brokers live.
| Method | What it does | Both brokers live | Source |
|---|---|---|---|
| Definitions export and import | Copies users, virtual hosts, queues, exchanges, bindings and runtime parameters as a JSON file. It moves the topology, not the messages. | Topology only | RabbitMQ docs |
| Federated queues | A federated queue pulls messages from a queue on another broker. In the blue-green guide, consumers on the new cluster receive messages still published to the old one. | Yes | RabbitMQ docs |
| Federated exchanges | Links an exchange to its counterpart on another broker so messages published on one side also reach the other. | Yes | RabbitMQ docs |
| Shovel | A client that consumes from a source queue or exchange on one broker and republishes to a destination on another. The blue-green guide uses it to move any backlog. | Yes | RabbitMQ docs |
| Blue-green deployment | The documented sequence: new cluster, import definitions, configure federation, move consumers, move producers, drain, decommission. | Yes | RabbitMQ docs |
Sources: RabbitMQ documentation, linked in each row; checked 30 Oct 2026.
How the methods were chosen: only methods the RabbitMQ project documents are listed. Anything the project does not describe, such as a custom proxy or a client-side dual write, is left out of the table and decided in the design review instead.
What Amazon MQ restricts during a move
Amazon MQ supports the shovel and federation plugins, but with conditions that depend on whether the broker is public or private. These conditions decide which side defines the link, so they are checked before the design is signed off.
| Area | What AWS documents | What it means for the move |
|---|---|---|
| Plugins | The supported list is management, shovel, federation, consistent hash exchange, OAuth 2.0, LDAP, HTTP, SSL certificate, aws, JMS topic exchange and Prometheus (AWS docs). | Every plugin the applications use must exist on the new cluster. The list is the starting point for the plugin parity check. |
| Shovel | Only dynamic shovels are supported, not static ones. A shovel cannot be configured when its destination is a private broker (AWS docs). | A shovel defined on the new cluster connects to the Amazon MQ broker as a client, so the new cluster's own settings apply. Network reachability between the two is a design item. |
| Federation | Federation cannot be configured if the upstream queue or exchange is in a private broker. It works between public brokers, or with a public upstream and a private downstream (AWS docs). | The link direction and the broker's public or private setting decide which side can define it. The design review confirms the option for the broker in question before anything is built. |
Sources: AWS Amazon MQ documentation, linked in each row; checked 30 Oct 2026.
Check these before you start
Most surprises in a broker move come from plugins, client addresses and credentials, not from the messages. Five checks cover them.
- Plugin parity.
- Compare the plugins the applications use with the Amazon MQ list above and with the design for the new cluster.
- Client switch.
- The RabbitMQ blue-green guide moves consumers by reconfiguring a load balancer or the consumer applications (RabbitMQ docs). Plan how each client finds the new broker.
- Definitions are secrets.
- Exported definitions contain user password hashes and hashing function information (RabbitMQ docs), so the file is handled like a credential.
- Version gap.
- The blue-green method also covers moves between versions that cannot be upgraded in place, which the RabbitMQ docs name as one of its uses (RabbitMQ docs).
- Broker type.
- This guide covers Amazon MQ for RabbitMQ. An Amazon MQ for ActiveMQ broker is a different engine, so moving it to RabbitMQ is an application review as well as a data move.
Sources: RabbitMQ and AWS documentation, linked above; checked 30 Oct 2026.
The migration in phases
The new cluster is built and linked while Amazon MQ keeps serving traffic. Nothing moves until the link is verified, and each phase leaves both brokers live.
Review
Queues, exchanges, plugins, connection counts, clients, and whether the Amazon MQ broker is public or private.
Build the new cluster
Node count, plugins and version are set by the design. The exported definitions are imported so queues, exchanges and bindings match.
Link both brokers
Federation or shovel carries messages between the old and new broker, in the direction the Amazon MQ conditions allow.
Move consumers, then producers
Consumers connect to the new cluster first. Producers follow in stages, and shovels drain any backlog on the old queues.
Decommission
The old broker is retired once it has no connections and no messages, and the new side is verified.
The timeline and runbook for Amazon MQ migrations are written in the design step, from your data volume, traffic and change windows, and you approve them before the cutover is scheduled.
Cutover and rollback
Consumers move first and producers second, so the new cluster serves reads before it takes writes. The old broker stays in place until the new side is verified, and the RabbitMQ docs describe federated queues as moving consumers without losing messages.
- Consumers reconnect to the new cluster. Federated queues deliver messages that are still published to Amazon MQ.
- Producers switch to the new cluster in stages, watching queue depths and error rates on both sides.
- Shovels move any remaining backlog from the old queues.
- The old broker shows no connections and empty queues, and traffic on the new side is normal. Only then is it decommissioned.
Source: the RabbitMQ blue-green guide (RabbitMQ docs); checked 30 Oct 2026.
How OSStend runs the move
We run the migration and own the cutover, with both sides live until you switch. Zero downtime is the plan.
RabbitMQ hosting from OSStend is operated by AceMQ engineers, who have run RabbitMQ for customers for 10+ years. The new cluster can sit in our cloud, your cloud account or your data center. The limits that usually start this conversation are set out on the Amazon MQ alternatives page and recorded with their sources in the managed service limits index, and the migration hub lists the other moves.
IBM MQ to RabbitMQ, without replacing the core
American National Insurance moved its Windows-based IBM MQ queues to RabbitMQ on Kubernetes first, with both brokers live until behaviour was validated, while IBM MQ licensing costs were progressively removed. Use case on acemq.com.
AceMQ engagementA Kafka CDC pipeline it did not need
American National replaced a Kafka-based change data capture pipeline for SQL Server with standalone Debezium publishing to RabbitMQ, removing Kafka clusters and Kafka Connect the use case did not need. Use case on acemq.com.
AceMQ engagementAceMQ names a customer only with that customer's written permission, so most of this record stays anonymous. Ask for a similar workload in the 45-minute review.
When this move fits, and when it does not
The move pays off when a plugin, node count or link to a broker outside AWS is the reason for leaving. It does not when Amazon MQ already covers the workload.
Contact us if
- ✓The workload needs more nodes, other plugins or links to brokers in a data center or another cloud.
- ✓The estate runs in more than one cloud and one team should do the move.
- ✓Messages must keep flowing during the move, with no planned pause.
Not a fit if
- ✕The workload fits Amazon MQ and AWS: staying avoids a migration.
- ✕You want to run the migration yourself: the RabbitMQ methods above work without OSStend.
- ✕The broker is ActiveMQ and the applications cannot change.
Questions about moving off Amazon MQ
Can Amazon MQ for RabbitMQ be migrated without downtime?
Yes, in principle. The RabbitMQ project documents a blue-green method in which federation and shovel carry messages while both clusters stay live. Which link types Amazon MQ allows depends on whether the broker is public or private (AWS docs, checked 30 Oct 2026).
Are messages lost during a blue-green move?
The RabbitMQ docs say federated queues let consumers move to the new cluster without disrupting consumption or losing messages (RabbitMQ docs, checked 30 Oct 2026). Persistent queues and publisher confirms are still reviewed in the design.
Do the definitions include messages?
No. Definitions hold users, virtual hosts, queues, exchanges, bindings and runtime parameters. Messages are moved by federation or shovel (RabbitMQ docs, checked 30 Oct 2026).
Related pages
- RabbitMQRabbitMQ hosting in your cloud or on premiseRabbitMQ hosting for clusters a plan cannot hold: federation and shovel, the plugins you need and upgrades on your change calendar, run 24/7 by engineers.
- AlternativeAmazon MQ alternatives for RabbitMQ in 2026Amazon MQ alternatives for RabbitMQ: clusters beyond three nodes, the plugins you need, upgrades on your calendar and more than one cloud. Limits sourced.
- Research · Original dataManaged service limits index, checked 9 Nov 2026A managed service limits index: 13 documented limits of Amazon MQ, MSK, ElastiCache, Neon, Supabase and Instaclustr, each with its source link and date.
- OverviewZero-downtime migration, owned by our engineersZero-downtime migration guides for Amazon MQ, RDS, Redis, Atlas, Oracle, MySQL, Confluent and Elasticsearch, using each engine's documented method.
Plan your Amazon MQ move with an engineer
45 minutes with an engineer: the current setup is reviewed against the documented method and the cutover is sketched. No ticket queue.
- ✓A design sketch for your workload, not a plan menu
- ✓A zero-downtime cutover path for what you run today
- ✓A price for that design, after the review