Compare Encore Pub/Sub, Amazon SQS and SNS, Google Cloud Pub/Sub, Amazon MQ, NATS JetStream, BullMQ on Redis and Kafka as replacements for a self-run RabbitMQ cluster, by routing model, delivery, ordering and what you still operate.

RabbitMQ is the broker many backends started with: exchanges and bindings for routing, acknowledgments and dead-letter exchanges for retries, and AMQP clients in every language. Running it is the part that ages badly.
A cluster needs quorum queues for durability since RabbitMQ 4.0 removed classic queue mirroring, memory and disk alarms that block publishers, Erlang and plugin upgrades, and someone who understands why a queue with a million unacked messages slowed the whole node.
The alternatives below split by need: RabbitMQ's routing model, or only reliable queues and fan-out between services. For head-to-heads see RabbitMQ vs SQS and Kafka vs RabbitMQ.
Encore brings infrastructure from code to TypeScript and Go backends, and most RabbitMQ setups, a work queue per job type and a fanout exchange per event, map onto Pub/Sub topics and subscriptions declared next to the code that uses them. Each subscription gets its own retry policy with backoff and a dead-letter queue, a maxConcurrency limit per instance, and optional ordering by attribute. encore run runs Pub/Sub in memory, so there is no local broker to start, and each publish and subscriber call appears in the local trace. On deploy to your own AWS account Encore provisions SNS and SQS, or Google Cloud Pub/Sub on GCP, and provisions an IAM role for the service, or a service account on GCP.
Run in your terminal to get started locally.
// images/jobs.ts
import { Topic, Subscription } from "encore.dev/pubsub";
export interface ImageUploaded {
imageId: string;
key: string;
}
// Replaces a fanout exchange: every subscription gets its own copy.
export const imageUploaded = new Topic<ImageUploaded>("image-uploaded", {
deliveryGuarantee: "at-least-once",
});
// Replaces a work queue with prefetch and a dead-letter exchange.
const _thumbs = new Subscription(imageUploaded, "make-thumbnails", {
maxConcurrency: 8,
ackDeadline: "60s",
retryPolicy: { minBackoff: "10s", maxBackoff: "10m", maxRetries: 10 },
handler: async ({ imageId, key }) => {
// Resize and store thumbnails. Idempotent: overwrite the same keys.
},
});
const _moderation = new Subscription(imageUploaded, "moderate", {
handler: async ({ imageId }) => {
// Call the moderation API.
},
});
What Encore does not replace is AMQP itself: there are no exchanges with arbitrary binding keys, no RPC over reply-to queues and no AMQP clients for other systems to connect with. If outside systems publish to your RabbitMQ over AMQP, keep a broker for that boundary.
A coding agent can add a new consumer as a subscription in the same change as the code it needs and test it under encore run, without editing a definitions file or creating queues and bindings by hand.
Best fit: TypeScript or Go backends that use RabbitMQ for work queues and fan-out between their own services and want those queues as managed SNS and SQS or Pub/Sub in their own AWS or GCP account.
SQS queues with visibility timeouts and redrive policies cover work queues; SNS topics with SQS subscriptions and filter policies cover fanout and topic-style routing. There is nothing to size or patch, and per the SQS quotas standard queues have nearly unlimited throughput. You trade RabbitMQ's push delivery for polling and its routing flexibility for filter policies, and you wire queues, topics, subscriptions, DLQs and IAM yourself, typically in Terraform.
Best fit: AWS teams that want managed queues and fan-out and are fine provisioning them as infrastructure.
Pub/Sub topics with pull or push subscriptions, filters, ordering keys, dead-letter topics and retry policies cover the same ground on GCP, and push subscriptions fit Cloud Run services that only scale on requests. See Pub/Sub push vs pull for the delivery choice.
Best fit: GCP teams replacing a self-run broker with a managed service.
Amazon MQ runs RabbitMQ brokers for you, so applications keep AMQP, exchanges and existing clients. AWS handles provisioning and patching during maintenance windows, while you still choose instance sizes, plan for broker limits and handle major version upgrades. It is the right move when the problem is operating RabbitMQ rather than RabbitMQ itself.
Best fit: teams that depend on AMQP features or third-party AMQP clients and want to stop running the servers.
NATS is a small, fast broker with subject-based addressing and wildcards; JetStream adds persistence, consumers with acknowledgments and redelivery limits. A three-node cluster is far lighter than RabbitMQ, and it runs well at the edge. You still operate it, and the client ecosystem is smaller than AMQP's.
Best fit: teams that want a self-hosted broker with less operational weight, especially for IoT or edge messaging.
For Node.js backends whose RabbitMQ is used as a job queue, BullMQ on Redis or Valkey gives delayed jobs, retries with backoff, rate limiting and repeatable jobs in a library. Durability depends on Redis persistence and memory, so a large backlog needs memory headroom, as discussed in BullMQ alternatives.
Best fit: Node.js apps that need background jobs and already run Redis.
Kafka replaces RabbitMQ when you need a retained, replayable log: event sourcing, change data capture, stream processing, or many consumers reading at their own pace. It is a different model, with consumer groups and partitions instead of per-message acknowledgment and routing, and more to run, as covered in Kafka alternatives.
Best fit: event streaming workloads that need retention and replay.
If the brokers are the problem and AMQP is a requirement, move to Amazon MQ. If you use RabbitMQ for work queues and fan-out between your own services, a managed queue and topic service removes the brokers, and Encore declares those topics and subscriptions in TypeScript or Go code. Pick NATS JetStream for a lighter self-hosted broker, BullMQ for Node.js job queues, and Kafka when retention and replay are the point.
If you use RabbitMQ for work queues and fan-out between your own services, a managed service such as SQS with SNS on AWS or Google Cloud Pub/Sub removes the brokers, and Encore declares those topics and subscriptions in code. If you rely on AMQP and exchange routing, Amazon MQ runs RabbitMQ for you. NATS JetStream is a lighter broker to self-host, BullMQ suits Node.js job queues on Redis, and Kafka fits when you need a retained, replayable log.
RabbitMQ 4.0 removed classic queue mirroring. Classic queues still work but only with a single replica, and replicated, highly available queues are quorum queues or streams. Clusters that relied on ha-mode mirroring policies have to migrate to quorum queues, which cannot be converted in place.
Deploy to your cloud on AWS and Google Cloud
without managing infrastructure
Encore lets developers and agents define application infrastructure directly in code, then automatically provisions it from local development to production in your AWS or GCP account.