RabbitMQ or Redpanda on Railway: which queue to pick

By Tracklayer · Updated 2026-11-20

Short answer: pick RabbitMQ if you need a job queue or message routing between services, where a consumer takes a message, acknowledges it and it is gone. Pick Redpanda if you need a Kafka-compatible log with retention and replay, or your code already uses a Kafka client. On Railway, Tracklayer’s templates run RabbitMQ 4.3.6 and Redpanda v26.2.4 as single, password-protected nodes with a volume.

Deploy RabbitMQ on Railway

How the two differ

RabbitMQ template Redpanda template
Protocol AMQP 0-9-1 and 1.0; plugins for MQTT, STOMP and streams Kafka API, with Schema Registry and HTTP Proxy built in
Model Exchanges and routing keys into queues, with acknowledgements, retries and dead-letter exchanges Topics with retention and replay
Typical uses Background jobs (for example Celery, Spring AMQP, MassTransit, NestJS), fan-out, absorbing bursts of webhooks Event streaming, change data capture, log ingestion, replay
Web UI Management UI with its own login Redpanda Console behind a Caddy basic-auth gateway
Version 4.3.6 v26.2.4 (Console v3.12.0)

If you only need to run background work for a web app, RabbitMQ is usually the smaller step. If several consumers need to read the same events independently, or you need to replay history, the Kafka model fits better.

Deploy RabbitMQ, step by step

These steps use Tracklayer’s RabbitMQ template.

  1. Deploy. No input is needed; the administrator password and Erlang cookie are generated.
  2. Sign in to the management UI with RABBITMQ_DEFAULT_USER and RABBITMQ_DEFAULT_PASS.
  3. Create a user per app. Under Admin, add a virtual host and a separate user for each application instead of sharing the administrator.
  4. Connect your service. In an app service in the same project, set RABBITMQ_URL to ${{RabbitMQ.RABBITMQ_URL}}. For Celery, set CELERY_BROKER_URL to the same reference.
  5. Declare durable queues. RabbitMQ 4.3 rejects non-durable queues that are not exclusive. Do not change RABBITMQ_NODENAME after the first deploy, or RabbitMQ starts with an empty database.

Deploy Redpanda, step by step

These steps use Tracklayer’s Redpanda template.

  1. Deploy. No input is needed; all passwords are generated.
  2. Sign in to Redpanda Console at the console-gateway domain with BASIC_AUTH_USERNAME and BASIC_AUTH_PASSWORD.
  3. Create a topic. Automatic topic creation is off by default.
  4. Connect your service. Set KAFKA_BROKERS to ${{redpanda.KAFKA_BOOTSTRAP_SERVERS}}, plus KAFKA_SASL_MECHANISM, KAFKA_SASL_USERNAME and KAFKA_SASL_PASSWORD from the same service. Use security protocol SASL_PLAINTEXT; traffic stays on the private network.
  5. Add least-privilege users. Create a separate user per application with rpk security user create in the redpanda service shell, and grant ACLs instead of sharing the superuser.
Deploy Redpanda on Railway

What it costs

Railway bills by usage. Both templates state their figures as follows.

For Redpanda, REDPANDA_MEMORY (default 1G) and retention length are the main cost levers.

Limits to know

Other options

For caching and simple queues backed by Redis, see the Redis and cache templates. To run background automation without writing consumers, see n8n in queue mode. All guides are listed at /guides/.

FAQ

How do I log in to the management UI?

RabbitMQ: open the service's public domain and sign in with RABBITMQ_DEFAULT_USER (default admin) and RABBITMQ_DEFAULT_PASS from the Variables tab. Redpanda: open the console-gateway public domain and use BASIC_AUTH_USERNAME and BASIC_AUTH_PASSWORD from that service's variables. All passwords are generated.

Is it safe to connect from outside Railway?

Both templates keep the broker on Railway's private network by default. An optional TCP proxy exposes it, but the proxy adds no TLS, so message contents travel unencrypted. Redpanda's SCRAM login does not send the password itself, but the data is still unencrypted. Use the proxy only for data that is not sensitive, or run clients on the private network.

Do messages survive a redeploy?

Yes, both use a volume. For RabbitMQ, publish messages as persistent and use durable queues. A redeploy briefly stops the broker because a volume can be mounted by only one deployment at a time.

Are there licence terms to know?

RabbitMQ is under the Mozilla Public License 2.0. Redpanda core is under the Business Source License 1.1, which allows production use for your own applications but not offering Redpanda as a streaming or queuing service to third parties. Each version converts to Apache-2.0 four years after release. This is a summary, not legal advice.

Can I run a cluster?

No. Both templates are single nodes, so Redpanda runs with replication factor 1. Scale vertically. Multi-node clusters are outside the scope of both templates.