Azure Service Bus Alternatives: Eight Options and What Each One Lacks
The main Azure Service Bus alternatives fall into four groups. If you stay on Azure, Storage Queues, Event Hubs and Event Grid each cover part of what Service Bus does. If you want a broker that runs anywhere, the candidates are RabbitMQ, Apache Kafka, Apache Artemis and NATS. If you are changing clouds, the counterparts are Amazon SQS with SNS on AWS and Google Cloud Pub/Sub. And for Java services that already run Valkey or Redis, Redisson PRO keeps queues and topics in that data store.
None of them is a drop-in replacement. Service Bus bundles features that the alternatives split up or leave out: sessions that order messages per key and carry state, scheduled messages, duplicate detection, a dead-letter queue on every entity, deferral, transactions and SQL rules on subscriptions. What a move costs depends on how many of those your entities use, and on what your code is written against: the Azure SDK, JMS, a .NET framework such as MassTransit or NServiceBus, Dapr, or Azure Functions triggers. This guide works out both before it compares the options.
Redisson is our product. It is option 8, for JVM services only, and the section on it says where it does not fit. We checked the Service Bus details against Microsoft Learn and the other products against their own documentation in November 2026.
Why Teams Look for an Azure Service Bus Alternative
Two deadlines fall three months apart. On September 30, 2026, Microsoft retired three Service Bus client libraries: WindowsAzure.ServiceBus and Microsoft.Azure.ServiceBus for .NET, and com.microsoft.azure.servicebus for Java. They still run, but they no longer get official support or updates, and SBMP, the protocol WindowsAzure.ServiceBus used by default, can no longer be used at all. Moving to Azure.Messaging.ServiceBus or azure-messaging-servicebus means rewriting the send, receive and settle code against a different API, and a rewrite is a natural point to ask whether Service Bus should stay underneath it.
.NET teams that reach Service Bus through MassTransit have a second date. MassTransit v9 is a commercial product sold by Massient, and the project's announcement limits official maintenance of the open-source v8 to the end of 2026. Staying on a maintained MassTransit now means buying a licence, and leaving it means touching the messaging code anyway.
Some features decide the tier for you. The JMS 2.0 API, messages over 256 KB, private endpoints and Geo-Replication are available only on Premium, which is billed per messaging unit per hour whether messages flow or not. Standard adds per-operation charges, counted in 64 KB units, to an hourly base charge, and every Standard namespace is throttled at 1,000 credits per second. Basic has no topics, sessions, transactions or duplicate detection.
Service Bus also runs only in Azure. Its on-premises edition, Service Bus for Windows Server, reached the end of support on January 10, 2023. The emulator Microsoft now ships runs in Docker for development and testing, and Microsoft's emulator documentation says not to use it for production workloads because it has no SLA. A team that needs the same messaging in its own data center or on another cloud has to pick a different product.
Other teams run into limits of the messaging model. Service Bus has no message priority, caps a message lock at five minutes so that long jobs must keep renewing it, returns an abandoned message for redelivery at once with no delay, and keeps no history to replay. Our Azure Service Bus comparison goes through each of these.
Microsoft keeps investing in the service, though. Geo-Replication became generally available on Premium in December 2025 and on partitioned Premium namespaces in November 2026, and since May 1, 2026, Premium namespaces in regions with availability zones carry a 99.99% uptime SLA. For many teams on Azure, the right result of this evaluation is to stay and move to the current SDK.
First, Find Out What Your Code Depends On
The same Service Bus queue can sit behind very different code, and each kind moves differently. Count the services in each row before you compare products.
| Your services use | Typically found in | What a move means |
|---|---|---|
The Azure SDK: Azure.Messaging.ServiceBus, azure-messaging-servicebus, or the Python, JavaScript and Go packages | Services in any language | A rewrite of send, receive and settlement code against the new system's client. Complete, abandon, dead-letter and defer each need a counterpart, or a decision to do without |
| One of the three retired libraries, or SBMP | Older .NET Framework and Java services | A rewrite is due whatever you choose, so do it once, against the destination |
JMS 2.0 through azure-servicebus-jms on Premium, often with Spring Cloud Azure's JMS starter | Java and Spring Boot services | Configuration on another JMS provider, such as Apache Artemis or Redisson PRO. Service Bus JMS cannot receive from session-enabled entities, so JMS consumers do not depend on sessions |
| Spring Cloud Azure's Service Bus starter, Spring Messaging, Spring Integration or the Spring Cloud Stream binder | Spring Boot services | A different starter or binder. Spring Cloud Stream binders exist for RabbitMQ, Kafka and Google Pub/Sub, and Redisson PRO ships one |
| MassTransit, NServiceBus, Wolverine or Dapr | .NET services, and Dapr in any language | Mostly transport configuration, within the limits described below |
| Azure Functions Service Bus triggers or Logic Apps connectors | Serverless consumers | A trigger for the new source. Storage Queues, Event Hubs and Event Grid have Functions triggers. Kafka bindings run on the Flex Consumption, Elastic Premium and Dedicated plans, and RabbitMQ bindings are fully supported only on Elastic Premium and Dedicated |
| A generic AMQP 1.0 client | Devices and less common languages | RabbitMQ 4.x and Artemis speak AMQP 1.0, but addressing, settlement and Service Bus's own management operations differ, so test before assuming the client works unchanged |
An inventory is mostly a search of code and configuration:
- package references to
Azure.Messaging.ServiceBus,Microsoft.Azure.ServiceBusandWindowsAzure.ServiceBusin .NET, and imports fromcom.azure.messaging.servicebusandcom.microsoft.azure.servicebusin Java; - Spring properties under
spring.cloud.azure.servicebusandspring.jms.servicebus; - MassTransit's
UsingAzureServiceBus, NServiceBus'sAzureServiceBusTransport, and Dapr components of typepubsub.azure.servicebus.queuesorpubsub.azure.servicebus.topics; - Service Bus trigger bindings in Azure Functions;
- the namespace itself, exported from Bicep, ARM, Terraform or
az servicebus: every queue and subscription with sessions, duplicate detection, auto-forwarding or a non-default lock duration, and every subscription rule, because each of those settings carries behavior a move has to replace (see the feature table).
If MassTransit, NServiceBus, Wolverine or Dapr Sits in Between
Many services, especially on .NET, never call the Service Bus SDK directly. They use a framework that supports several transports, and for them the realistic alternatives are the transports that framework offers:
- MassTransit supports RabbitMQ, Azure Service Bus, Amazon SQS, ActiveMQ and a SQL transport on PostgreSQL or SQL Server, and treats Kafka and Event Hubs as riders, a separate model from its broker transports. Version 9 needs a commercial licence.
- NServiceBus, a commercial product from Particular Software, supports Azure Service Bus, Azure Storage Queues, Amazon SQS, RabbitMQ, SQL Server, PostgreSQL, IBM MQ and MSMQ.
- Wolverine, under the MIT licence, supports RabbitMQ, Azure Service Bus, Amazon SQS and SNS, Google Pub/Sub, Kafka, Pulsar, NATS, Redis, MQTT and several databases, among others.
- Dapr's publish and subscribe API has stable components for Service Bus queues and topics, Event Hubs, RabbitMQ, Kafka, Redis Streams, AWS SNS/SQS, Google Pub/Sub and Pulsar, plus a beta component for NATS JetStream, so a switch starts with a different component file.
Behind the common API, each transport still decides what the features do. Where Service Bus has a native feature, such as scheduled delivery, subscription rules or dead-lettering, a framework like MassTransit uses it. On a transport without that feature, the framework either emulates it with infrastructure of its own or does not offer it, so read the target transport's documentation for every Service Bus feature your entities use before you count the switch as configuration only. The SQL transports in MassTransit and NServiceBus are worth a look at modest volumes, because they turn a database you already run into the broker.
The Service Bus Features a Move Has to Replace
Most Service Bus features are switched on per queue, topic or subscription, so the namespace export from the inventory tells you which of these you use. The table says what each one does and where else it exists.
| Service Bus feature | What it does | Where else you get it |
|---|---|---|
| Sessions | Messages with the same session ID go, in order, to the receiver holding that session's exclusive lock, and the session can store state. Standard and Premium only | Per-key order: SQS FIFO message groups, Kafka partition keys, Pub/Sub ordering keys, Artemis message groups and Redisson PRO message groups. RabbitMQ orders a whole queue with a single active consumer. Session state exists nowhere else; keep it in a database or cache |
| Scheduled messages | A message stays invisible until its scheduled enqueue time and can be cancelled by sequence number. Microsoft states no maximum horizon | Artemis (a scheduled delivery property and JMS 2.0 delivery delay), RabbitMQ 4.4 quorum queues, NATS 2.12 and later, and Redisson PRO, with a delay per message. Storage Queues can hide a new message for up to 7 days, and SQS delays it by up to 15 minutes. Kafka and Pub/Sub have nothing; Google points to Cloud Tasks |
| Duplicate detection | Drops a message whose MessageId was already seen within a window of 20 seconds to 7 days, 10 minutes by default. Standard and Premium only | SQS FIFO queues (a 5-minute interval), Artemis (a duplicate ID property), NATS JetStream (a 2-minute window by default) and Redisson PRO (by ID or payload hash). Kafka's idempotent producer removes duplicates caused by producer retries. Storage Queues, RabbitMQ queues and Pub/Sub publishing have none built in |
| Dead-letter queue | Every queue and subscription has one that cannot be deleted. Messages land there after MaxDeliveryCount deliveries, 10 by default, and code can dead-letter a message with a reason | Elsewhere you configure it: SQS redrive policies, Pub/Sub dead-letter topics, RabbitMQ dead-letter exchanges, Artemis dead-letter addresses, Event Grid dead-letter destinations and Redisson PRO's deadLetterQueueName. Without one, RabbitMQ, Artemis and Redisson PRO drop a message that is rejected or runs out of attempts. NATS has no dead-letter queue, only an advisory event; Kafka has one only for share groups, since 4.4; Storage Queues leave it to the consumer |
| Deferral | A receiver sets a message aside and fetches it later by sequence number | No direct equivalent among the alternatives here. Move the message to a holding queue or a store and fetch it from there |
| Transactions | Send, complete, abandon, dead-letter and defer in one transaction, across entities through send-via, with a 2-minute timeout | Artemis (JMS local transactions and XA), Kafka (transactions across Kafka topics only) and NATS (atomic batch publishing, on the publish side only). SQS, SNS, Pub/Sub and Storage Queues have none, the RabbitMQ JMS Client's local transactions do not guarantee atomicity, and Redisson PRO documents no transaction across operations |
| Subscription rules and auto-forwarding | SQL and correlation filters on message properties, evaluated by the broker; actions that rewrite properties; forwarding from one entity to another | RabbitMQ exchanges and bindings, SNS subscription filter policies, Pub/Sub filters (attributes only, fixed when the subscription is created), Event Grid filters, NATS subject wildcards, and Artemis filters and diverts. Kafka has no broker-side filtering. Redisson PRO's Reliable Fanout filters run in the publisher and cannot rewrite a message |
| Lock duration | One minute by default and five at most; longer work renews the lock | SQS visibility timeout, up to 12 hours; Storage Queues, up to 7 days; Pub/Sub ack deadlines of 10 to 600 seconds, which the client libraries extend; NATS's acknowledgement wait, 30 seconds by default, which a consumer can extend while it works; Redisson PRO's visibility timeout, with no documented maximum. RabbitMQ and Artemis hold an unacknowledged message while the consumer stays connected, within RabbitMQ's 30-minute consumer timeout |
| Message size | 256 KB on Basic and Standard; up to 100 MB on Premium | Storage Queues 64 KB; SQS 1 MiB; SNS 256 KiB by default and 1 MiB at most, or up to 2 GB through Amazon's extended client libraries; Event Hubs 1 MB (256 KB on Basic, 20 MB on Dedicated); Event Grid 1 MB; Pub/Sub 10 MB. Brokers you run are configurable; RabbitMQ allows 16 MiB by default |
| JMS 2.0 | Premium only, through azure-servicebus-jms; Standard has a queue-focused JMS 1.1 subset | Apache Artemis (Jakarta Messaging 3.1 and JMS 2.0, with XA), Redisson PRO (JMS 2.0, Jakarta Messaging 3.0 and 3.1), the RabbitMQ JMS Client (with gaps), ActiveMQ Classic (partial) and SQS (JMS 1.1, queues only). Pub/Sub and NATS have no JMS provider, and Confluent's JMS 1.1 client for Kafka is deprecated |
Dead-lettering is safe by default on Service Bus: a poison message always ends up somewhere you can inspect it. On RabbitMQ quorum queues, Artemis and Redisson PRO, the same message is deleted once it runs out of attempts unless you name a destination, so set one for every queue you migrate. And session state has no counterpart anywhere on this list, so code that stores state in sessions needs a new home for it whichever option you pick.
The Eight Alternatives
The options run from the smallest move, another Azure service, to the largest, another cloud or a different kind of data store. Each closes with the situation in which it is the wrong pick. Review sites also list MuleSoft Anypoint, IBM App Connect, webMethods and Oracle Service Bus as Service Bus alternatives, because they file Azure Service Bus under enterprise service bus software. Those are integration platforms that sit above messaging, so they are left out here, and so are NServiceBus and MassTransit, which run on top of a broker (see above).
1. Stay on Azure: Storage Queues, Event Hubs or Event Grid
Microsoft's comparison of its messaging services describes Service Bus as enterprise transactional messaging, Event Hubs as big data streaming and Event Grid as reactive event routing. The other two, and Storage Queues, each cover part of what Service Bus does:
- Azure Storage Queues are simple work queues inside a storage account. A queue can grow to the account's capacity, a message is at most 64 KB and can be kept indefinitely, and a new message can be hidden for up to 7 days, which works as a basic delay. There are no topics, no ordering guarantee, no duplicate detection, no sessions and no automatic dead-lettering: consumers check each message's dequeue count, and the Azure Functions queue trigger moves a message to a queue with a
-poisonsuffix after five failed attempts. Microsoft's queue comparison lists the differences row by row. - Azure Event Hubs is a partitioned log. It keeps events for 1 day on Basic, 7 on Standard and 90 on Premium and Dedicated, lets several consumer groups read the same stream (one on Basic), orders events within a partition, and has a Kafka endpoint on Standard and above. Consumers track their position with checkpoints instead of settling messages, so there is no per-message retry, dead-lettering or scheduling.
- Azure Event Grid routes discrete events to handlers by push. Its namespaces add pull delivery with queue-like semantics: a consumer acknowledges, releases (optionally after a delay) or rejects each event, and a rejected event goes to a dead-letter destination if one is configured, or is dropped. Namespace topics keep events for up to 7 days, events are capped at 1 MB, and namespaces also include an MQTT broker.
Staying on Azure keeps identity, networking, monitoring and Functions triggers as they are, and Microsoft's own documentation covers the choice between these services well.
Wrong for you if: you are leaving Azure, or your entities rely on sessions, duplicate detection, transactions or automatic dead-lettering. Storage Queues have none of them, and Event Hubs and Event Grid do not offer them in the form Service Bus does.
2. RabbitMQ
RabbitMQ's model maps onto Service Bus with little translation: queues with per-message acknowledgements, plus exchanges and bindings that route messages to them and take over from topics, subscriptions and their rules. Quorum queues replicate messages across nodes and dead-letter through a dead-letter exchange after a delivery limit of 20 by default. Version 4.3 added delayed retries and up to 32 priority levels on quorum queues, and 4.4 added delayed delivery. Streams add a log that can be replayed. Since 4.0 RabbitMQ has spoken AMQP 1.0 natively, alongside AMQP 0-9-1, MQTT and STOMP. That still does not make it a drop-in target for Service Bus clients: the Service Bus SDKs rely on Service Bus's own operations for things such as lock renewal and scheduling, and RabbitMQ has its own address format, so plan on client changes either way.
Azure has no first-party RabbitMQ service. You run it on AKS with the RabbitMQ Cluster Kubernetes Operator, or rent it: CloudAMQP runs managed RabbitMQ clusters in Azure regions, sells them through Azure Marketplace and supports Azure Private Link. The community supports each release series for a short time, 4.4 until July 31, 2027, and Broadcom sells longer support with Tanzu RabbitMQ. MassTransit, NServiceBus and Wolverine all have RabbitMQ transports, while the Azure Functions RabbitMQ bindings are fully supported only on the Elastic Premium and Dedicated plans.
Against Service Bus it lacks sessions with state (a single active consumer orders a whole queue, not each key), duplicate detection on queues, and deferral. JMS comes through the RabbitMQ JMS Client, without selectors on queues, without XA, and with local transactions that its documentation says do not guarantee atomicity. RabbitMQ alternatives covers the move in the other direction, and ActiveMQ vs RabbitMQ goes deeper into RabbitMQ's support calendar and operations.
Wrong for you if: you want Microsoft to run the broker, or your entities depend on sessions, duplicate detection or deferral.
3. Apache Kafka, Confluent Cloud or the Event Hubs Kafka Endpoint
Kafka belongs on the list when the reason for leaving is streaming: replay, long retention or stream processing, none of which Service Bus does. It is a partitioned log, so the model changes. Order holds within a partition, consumers commit offsets instead of settling messages, and the broker schedules, prioritizes and deduplicates nothing beyond what the idempotent producer prevents. Share groups, production-ready since Kafka 4.2, add per-record acknowledgement for queue-style consumption, and Kafka 4.4 added dead-letter queues for them, which narrows the gap for plain work queues.
On Azure there are three ways to get it. Event Hubs has a Kafka endpoint on its Standard, Premium and Dedicated tiers for Kafka clients from version 1.0 on, with Kafka Streams and Kafka transactions still in preview on Premium and Dedicated. Confluent Cloud is sold as an Azure Native ISV Service, developed and managed by Microsoft and Confluent, and Confluent has been part of IBM since March 2026. Or you run Kafka yourself on AKS. MassTransit supports Kafka and Event Hubs as riders rather than as transports. Apache Kafka alternatives and Is Kafka a Message Queue? go further into the difference between a queue and a log.
Wrong for you if: the workload is a work queue that needs scheduled delivery, delayed retries, broker-side duplicate detection or filtering, and no consumer ever needs to re-read old messages.
4. Apache Artemis and Other JMS Brokers
Among brokers you run yourself, Apache Artemis keeps the most Service Bus features. It implements all of Jakarta Messaging 3.1 and of JMS 2.0 and 1.1, XA included, and accepts AMQP 1.0, MQTT, STOMP and OpenWire clients as well as its own core protocol. It schedules delivery through a message property or JMS 2.0 delivery delay, detects duplicates through a duplicate ID property backed by a persisted cache of 20,000 IDs by default, sends each message group to a single consumer, and moves a message to a dead-letter address after 10 delivery attempts by default. If no dead-letter address is configured, it discards that message, and it creates dead-letter resources automatically only when told to. Artemis has been an Apache project of its own since November 2025, and 2.57.0, released on September 9, 2026, fixes security issues present in earlier releases, so start there or later.
There is no Azure-managed Artemis: you run it on AKS or virtual machines, and Red Hat sells commercial support for it as AMQ Broker. Clients outside the JVM connect over AMQP 1.0, MQTT or STOMP. Code written against azure-servicebus-jms gains features on Artemis: XA and noLocal work, and JMS message groups give it per-key ordering, which Service Bus JMS cannot offer because it cannot receive from session-enabled entities. The commercial JMS brokers, IBM MQ, TIBCO EMS and Solace, are compared in IBM MQ alternatives, and ActiveMQ Classic, whose JMS 2.0 support is partial, in ActiveMQ alternatives.
Wrong for you if: nobody on the team will run a broker, or you want a service someone else operates on Azure.
5. NATS JetStream
NATS is a CNCF messaging system that ships as a single server binary under the Apache 2.0 licence, with clients in many languages, .NET included. JetStream, its persistence layer, stores streams on disk or in memory with replicas, and its consumers acknowledge each message, within an acknowledgement wait of 30 seconds by default that a consumer can extend while it works. A stream can drop duplicates by the Nats-Msg-Id header within a window of two minutes by default, and can run as a work queue that deletes each message once it is acknowledged. Since 2.12 it can hold a message for delivery at a later time and publish a batch of messages atomically, and 2.14 added recurring schedules. The current release is 2.15, from September 17, 2026.
The gap that matters most to a Service Bus user is dead-lettering. JetStream has no dead-letter queue: by default it redelivers a message without limit, and once a consumer's maximum delivery count is reached, the server publishes an advisory event that your own code has to turn into a dead-letter stream. It has no JMS, no transactions beyond batch publishing, and nothing like sessions with state. Synadia, the company behind NATS, runs it as a multi-cloud service, Synadia Cloud, or operates it for you in your own cloud account, Azure included, as Synadia Platform; otherwise you run it on AKS. Wolverine has a NATS transport, and Dapr a beta JetStream component.
Wrong for you if: you need the broker to dead-letter, run transactions or speak JMS, or you want a service Microsoft operates.
6. Amazon SQS and SNS: the AWS Equivalent
The AWS equivalent of Azure Service Bus is a pair of services. Microsoft's map of AWS and Azure messaging pairs Amazon SQS with Queue Storage or Service Bus queues, SNS with Service Bus topics or Event Grid, and Amazon MQ with Service Bus for JMS applications. Read in reverse, Service Bus queues become SQS queues, a topic with subscriptions becomes an SNS topic that fans out to one SQS queue per subscriber, and code that needs a broker protocol such as JMS or AMQP goes to Amazon MQ, which runs ActiveMQ and RabbitMQ. Feature by feature:
- Sessions map to FIFO queues, which keep strict order within a message group and return no more messages from a group until the ones in flight are deleted or become visible again. Session state has no counterpart.
- Scheduled messages map to delay queues and message timers, which hold a message for up to 15 minutes. Anything longer needs a scheduler in front, such as a one-time EventBridge Scheduler schedule that sends to the queue.
- Duplicate detection maps to FIFO deduplication, by an explicit ID or a SHA-256 hash of the body, within a 5-minute interval.
- The dead-letter queue becomes a second queue named in a redrive policy, with
maxReceiveCountin place ofMaxDeliveryCount, configured for each source queue. - The message lock becomes the visibility timeout: 30 seconds by default, up to 12 hours, and extendable while a consumer works.
- Subscription rules become SNS subscription filter policies, up to 200 per topic.
- The size limit moves from 256 KB, or 100 MB on Premium, to 1 MiB on SQS, and to 256 KiB by default on SNS, raisable to 1 MiB.
- Transactions and deferral have no equivalent, and JMS shrinks to the Amazon SQS Java Messaging Library, which implements JMS 1.1 for queues only.
AWS publishes a migration pattern for .NET queue code, and MassTransit, NServiceBus, Wolverine and Dapr all support SQS. Amazon SQS vs Reliable Queue goes through SQS in more detail, and for JVM services on AWS, JMS on AWS without a broker covers Redisson PRO on ElastiCache or MemoryDB.
Wrong for you if: your entities depend on transactions, deferral or session state, or you schedule messages more than 15 minutes ahead and do not want a separate scheduler.
7. Google Cloud Pub/Sub: the Google Cloud Equivalent
On Google Cloud, Pub/Sub replaces both Service Bus topics and queues: a topic fans out to subscriptions, and the consumers of a subscription share its messages and acknowledge each one. A subscription can be pulled, pushed to an HTTPS endpoint or exported to BigQuery or Cloud Storage. Ack deadlines run from 10 to 600 seconds, ordering keys keep per-key order, attribute filters choose what a subscription receives, and once you attach a dead-letter topic, it takes a message after 5 to 100 failed deliveries. Exactly-once delivery is available for pull subscriptions within one region. Unlike Service Bus, Pub/Sub can keep messages for up to 31 days and let a subscription seek back to replay them.
It has no scheduled delivery, for which Google points to Cloud Tasks, no priority, no transactions, no deferral, and neither JMS nor AMQP. A filter cannot change after its subscription is created. Google Cloud also offers no managed RabbitMQ, so a broker protocol means running a broker on GKE or renting one. Pub/Sub is billed by data volume, counting each subscription's deliveries as well as the publishes. Google Pub/Sub vs Redis and Google Pub/Sub vs Kafka cover the service in depth, and Google Pub/Sub alternatives covers what else Google Cloud offers, Managed Service for Apache Kafka and Cloud Tasks among them.
Wrong for you if: you schedule messages, rely on session state or transactions, or need JMS.
8. Redisson PRO on Valkey or Redis
Redisson is a Java client for Valkey and Redis, and Redisson PRO, its commercial edition, runs messaging inside the Valkey or Redis deployment you already have, such as Azure Managed Redis, so no broker is added. It is for JVM services only. Reliable Queue takes the place of Service Bus queues, Reliable PubSub of topics and subscriptions, and Reliable Fanout routes one publish to several queues through filters. A JMS 2.0 and Jakarta Messaging 3.0 and 3.1 implementation sits on top, and its documentation states that it passes all TCK tests, so code on azure-servicebus-jms can keep its JMS API without a Premium namespace.
From the feature table above, it covers scheduled delivery with a delay per message, duplicate detection by message ID or payload hash with a window per message, delivery limits with a dead-letter queue, a visibility timeout with no documented maximum, and ordered processing by message group on Reliable PubSub, available since Redisson PRO 4.0.0. It adds three things Service Bus lacks: a priority from 0 to 9 on each message, retries with a delay through a negative acknowledgement, and replay, for which a subscription created with retainAfterAck() keeps acknowledged messages and can seek back to an earlier message ID or timestamp. Two defaults differ from Service Bus: the visibility timeout is 30 seconds, where a Service Bus lock lasts a minute, and no dead-letter queue exists until you name one. Set both when you create a queue:
RReliableQueue<Order> queue = redisson.getReliableQueue("orders");
queue.setConfig(QueueConfig.defaults()
.deliveryLimit(10) // MaxDeliveryCount
.visibility(Duration.ofMinutes(1)) // LockDuration
// Service Bus creates a dead-letter queue for every entity;
// here, without a name, exhausted messages are deleted
.deadLetterQueueName("orders-dlq"));
// Scheduled, deduplicated and prioritized in one call
queue.add(QueueAddArgs.messages(MessageArgs.payload(order)
.delay(Duration.ofMinutes(30)) // ScheduledEnqueueTime
.deduplicationById(order.id(), Duration.ofMinutes(10))
.priority(9))); // Service Bus has none
It does not replace session state, broker-side SQL rules with actions (Reliable Fanout filters run in the publisher), deferral, transactions across operations, Azure Functions triggers or clients outside the JVM. Messages live in memory, so provision memory for the largest backlog you expect, use an eviction policy that never removes messaging keys, and turn on high availability and persistence. On Azure Managed Redis, the append-only file is written once per second, and persistence cannot be combined with active geo-replication. By default, Redisson only logs a warning when a write is not confirmed within its sync timeout, so set syncFailureMode to THROW_EXCEPTION where the producer must see that failure. For the full comparison, with worker code, a map of Service Bus settings and a step-by-step migration, see Azure Service Bus vs Reliable Queue and Reliable PubSub.
The PRO licence includes 24x7 support with an SLA for the client, while Valkey or Redis is supported by whoever runs it, and a free trial runs against your own deployment.
Wrong for you if: any consumer runs outside the JVM or relies on an Azure Functions trigger, your entities depend on session state, SQL rules or transactions, or your backlogs can outgrow memory.
Azure Service Bus Alternatives Compared
| Option and model | Who runs it | Clients and protocols | JMS | Licence |
|---|---|---|---|---|
| Azure Service Bus Managed broker | Microsoft, on Azure | SDKs for .NET, Java, Python, JavaScript and Go; AMQP 1.0; HTTPS | JMS 2.0 on Premium; a JMS 1.1 subset on Standard | Cloud service |
| Azure Storage Queues Managed queue | Microsoft, on Azure | Azure Storage SDKs; REST | None | Cloud service |
| Azure Event Hubs Managed partitioned log | Microsoft, on Azure | AMQP; Kafka protocol; HTTPS | None | Cloud service |
| Azure Event Grid Managed event routing | Microsoft, on Azure | HTTP; MQTT | None | Cloud service |
| RabbitMQ Broker | You, on AKS or virtual machines, or CloudAMQP | AMQP 0-9-1 and 1.0, MQTT, STOMP; clients in many languages | RabbitMQ JMS Client, with gaps | MPL 2.0; Tanzu RabbitMQ is commercial |
| Apache Kafka Partitioned log | Event Hubs (Kafka endpoint), Confluent Cloud, or you | Kafka protocol; clients in many languages | None; Confluent's JMS 1.1 client is deprecated | Apache 2.0; Confluent is commercial |
| Apache Artemis Broker | You | AMQP 1.0, MQTT, STOMP, OpenWire, core protocol | Jakarta Messaging 3.1, JMS 2.0 and 1.1, with XA | Apache 2.0; Red Hat AMQ Broker is commercial |
| NATS JetStream Messaging system with persistent streams | You, or Synadia | NATS protocol; clients in many languages | None | Apache 2.0; Synadia's services are commercial |
| Amazon SQS and SNS Managed queues plus SNS fan-out | AWS, on AWS | AWS SDKs; HTTPS | JMS 1.1, queues only | Cloud service |
| Google Cloud Pub/Sub Managed publish/subscribe | Google, on Google Cloud | Client libraries in nine languages; gRPC; REST | None | Cloud service |
| Redisson PRO Java library; messages are kept in Valkey or Redis | You or your Valkey or Redis provider, such as Azure Managed Redis | Java only, over the Valkey and Redis protocol | JMS 2.0, Jakarta Messaging 3.0 and 3.1 | Commercial licence; Valkey and Redis have their own |
How to Choose
Read the rows in order and take the first one that fits. Many teams fit several, so make the call for each queue or topic separately.
| If this describes you | Look at |
|---|---|
| Your entities use session state, SQL rules with actions, deferral or transactions across entities, and you are staying on Azure | Service Bus. If the retired SDK or SBMP started the evaluation, move to the current SDK; if the bill did, check what Premium is paying for |
| You are moving to AWS | SQS with SNS, FIFO queues where order or duplicate detection matters, and a scheduler for delays over 15 minutes; Amazon MQ for JMS or AMQP brokers; Kinesis Data Streams or Amazon MSK where consumers replay; for JVM-only estates, Redisson PRO on ElastiCache or MemoryDB |
| You are moving to Google Cloud | Pub/Sub, which also replays retained messages, with Cloud Tasks for scheduled work; Redisson PRO on Memorystore if every client is on the JVM, with persistence on and the eviction policy set to noeviction |
| Simple, high-volume work queues on Azure, with no ordering, topics or duplicate detection | Azure Storage Queues |
| Stream processing, or replay where some readers are not on the JVM or the history will not fit in memory | Event Hubs on Azure, through its Kafka endpoint if your clients speak Kafka, or Kafka on Confluent Cloud or your own cluster |
| Replay of past messages, with every reader in Java or another JVM language and a retained history small enough for memory | Redisson PRO: a Reliable PubSub subscription created with retainAfterAck() keeps acknowledged messages and can seek back to a timestamp or message ID, while other subscriptions on the topic work like queues |
| All clients in Java or another JVM language, Valkey or Redis already running (Azure Managed Redis or your own), no need for XA, and a need for priority, delayed retries or JMS 2.0 without Premium | Redisson PRO, configured as in option 8 |
| JMS code that needs XA, or a JMS broker of your own that clients in several languages share | Apache Artemis, with Red Hat for commercial support |
| A broker that runs anywhere, with clients in many languages and flexible routing | RabbitMQ, on AKS or from CloudAMQP |
Is Kafka a Message Queue? goes further into logs versus brokers, and the JMS broker comparison lists the specification level each provider implements.
Azure Service Bus Alternatives: Frequently Asked Questions
What Is the Best Alternative to Azure Service Bus?
It depends on why you are leaving. To stay on Azure with simpler queues, use Azure Storage Queues; for streaming and replay, Event Hubs or Kafka; for a broker you can run anywhere, RabbitMQ, or Apache Artemis if your code uses JMS; on AWS, SQS with SNS; on Google Cloud, Pub/Sub. For Java services that already run Valkey or Redis, Redisson PRO keeps queues and topics there. None of them has sessions with state, and most make you configure the dead-lettering that Service Bus provides by default.
What Is the AWS Equivalent of Azure Service Bus?
Amazon SQS for queues and Amazon SNS for topics, usually with one SQS queue per subscriber behind each SNS topic. Microsoft's own map of AWS and Azure services pairs them this way, and pairs Amazon MQ with Service Bus for JMS applications. SQS FIFO queues give ordering per message group and a 5-minute deduplication interval, message delays stop at 15 minutes, and there are no transactions or session state.
What Is the Google Cloud Equivalent of Azure Service Bus?
Google Cloud Pub/Sub. Its topics fan out to subscriptions whose consumers acknowledge each message, with ordering keys, attribute filters and dead-letter topics, and it keeps messages for up to 31 days for replay. It has no scheduled delivery, for which Google points to Cloud Tasks, and Google Cloud offers no managed RabbitMQ.
Is There an Open-Source Alternative to Azure Service Bus?
Yes, several. RabbitMQ, under the MPL 2.0, has the closest broker model. Apache Artemis, under the Apache 2.0 licence, covers JMS, scheduled delivery, duplicate detection and message groups. NATS JetStream is the lightest to run, and Apache Kafka fits streaming. You run them yourself, on AKS or virtual machines, or pay a provider such as CloudAMQP, Confluent or Synadia to run them for you.
Can Azure Service Bus Run On-Premises?
Not in production. Service Bus for Windows Server, the on-premises edition, reached the end of support on January 10, 2023. The Service Bus emulator runs in Docker for development and testing, but Microsoft says not to use it for production because it has no SLA, and it supports neither JMS nor keeping data across container restarts. For production on premises, teams use a broker such as RabbitMQ or Apache Artemis.
Is Azure Service Bus Being Deprecated?
No. The service is supported and still gaining features, such as Geo-Replication on partitioned namespaces and a 99.99% SLA for Premium namespaces in regions with availability zones. What Microsoft retired on September 30, 2026 were three old client libraries and the SBMP protocol. Applications that still use them should move to the current Azure SDK, Azure.Messaging.ServiceBus or azure-messaging-servicebus.
Is Azure Service Bus the Same as Kafka?
No. Service Bus is a message broker: each message is settled on its own, retried, dead-lettered if it keeps failing, and removed once it is completed. Kafka, and Azure Event Hubs, which offers a Kafka endpoint, are partitioned logs that keep events for a retention period so that many consumers can read and replay them. Service Bus suits commands and work queues; Event Hubs and Kafka suit event streams.
Is RabbitMQ a Good Replacement for Azure Service Bus?
Often, for teams that want a broker they can run anywhere. Exchanges and bindings replace topics and subscription rules, quorum queues replicate messages and dead-letter them through a dead-letter exchange, and recent releases add delayed retries, delayed delivery and priorities. You give up sessions with state, duplicate detection on queues, deferral and a service run by Microsoft, although providers such as CloudAMQP run RabbitMQ on Azure for you.
Next Steps
If Redisson PRO is on your shortlist, Azure Service Bus vs Reliable Queue and Reliable PubSub maps Service Bus settings onto it and walks through a migration, and the Redis Java client for Azure Cache covers connecting Redisson on Azure. JMS messaging over Valkey and Redis shows the JMS API in use, the messaging documentation has the full configuration, and a free Redisson PRO trial runs against your own Valkey or Redis.
Similar articles
- Azure Service Bus vs Valkey & Redis Based Reliable Queue and PubSub
- Amazon SQS vs. Reliable Queue for Valkey and Redis
- Google Pub/Sub vs Redis for Messaging
- Google Pub/Sub Alternatives: Eight Options On and Off Google Cloud
- RabbitMQ Alternatives: Seven Options, and What Each One Breaks
- Apache Kafka Alternatives: 7 Options Compared
- ActiveMQ Alternatives: Seven Options and What Happens to Your JMS Code
- IBM MQ Alternatives: Eight Options, and What Moves Without a Rewrite
- JMS Messaging Over Valkey and Redis