Azure Service Bus

---
name: azure-service-bus
description: "Expert coverage of Azure Service Bus managed messaging service: queues, topics, subscriptions, sessions, dead-letter queues, duplicate detection, transactions, Premium tier, geo-disaster recovery, and SDK patterns. Use for \"Azure Service Bus\", \"Service Bus\", \"Service Bus queue\", \"Service Bus topic\", \"Service Bus subscription\", \"Service Bus session\", \"SessionId\", \"peek-lock\", \"PeekLock\", \"dead letter queue Service Bus\", \"DLQ Service Bus\", \"Service Bus Premium\", \"messaging unit\", \"MU\", \"Service Bus namespace\", \"geo-disaster recovery\", \"geo-replication Service Bus\", \"az servicebus\", \"Azure.Messaging.ServiceBus\", \"Service Bus filter\", \"SQL filter\", \"correlation filter\", \"Service Bus transaction\", \"message deferral\", \"scheduled message Service Bus\", \"auto-forwarding\". Do NOT use for general Azure architecture or non-messaging resource provisioning -- that's the `cloud-platforms` plugin."
license: MIT
---

# Azure Service Bus

This skill covers Azure Service Bus, Microsoft's fully managed enterprise message broker. Service Bus is a managed service with no user-facing version numbers. It includes:

- Queues (point-to-point, competing consumers, PeekLock/ReceiveAndDelete)
- Topics and subscriptions (pub/sub fan-out, SQL and correlation filters)
- Sessions (guaranteed FIFO per entity, session state, sequential convoy)
- Dead-letter queues (built-in per entity, application-level dead-lettering)
- Duplicate detection (message ID tracking with configurable window)
- Transactions (atomic operations across queue entities)
- Scheduled messages and message deferral
- Auto-forwarding (entity chaining within namespace)
- Premium tier (dedicated MUs, VNET, large messages up to 100 MB, geo-DR)
- SDKs (.NET, Python, Java, JavaScript) and Azure CLI management
- Monitoring (Azure Monitor metrics, Log Analytics, diagnostics)

## How to Approach Tasks

1. **Classify** the request:
   - **Architecture / design** -- Load `references/architecture.md` for namespace model, queues, topics, sessions, Premium tier, geo-DR
   - **Best practices** -- Load `references/best-practices.md` for tier selection, connection management, prefetch, session design, security
   - **Troubleshooting** -- Load `references/diagnostics.md` for throttling, DLQ analysis, session issues, connectivity, cost

2. **Gather context** -- Tier (Basic/Standard/Premium), partitioned vs non-partitioned, session-enabled, receive mode, connection method (connection string vs Azure AD)

3. **Recommend** -- Provide actionable guidance with Azure CLI commands, SDK code, Bicep templates, and monitoring queries.

## Core Architecture

### Namespace Model

A namespace is the scoping container: `<namespace>.servicebus.windows.net`. Contains queues, topics, and subscriptions. Premium namespaces are zone-redundant.

### Queues
Point-to-point FIFO delivery. PeekLock (default, at-least-once) or ReceiveAndDelete (at-most-once). Max delivery count default 10. Built-in DLQ at `<queue>/$deadletterqueue`.

### Topics and Subscriptions
Pub/sub fan-out. Up to 2,000 subscriptions per topic. Three filter types: Boolean, Correlation (most efficient), SQL (flexible, higher cost).

### Sessions
`SessionId` property groups messages for ordered per-entity processing. Consumer acquires exclusive session lock. Session state storage (256 KB Standard, 100 MB Premium).

### Premium Tier
Dedicated compute (1-16 MUs). VNET integration. Messages up to 100 MB (AMQP only). Geo-DR and geo-replication. Zone redundancy. JMS 2.0. Auto-scale.

## Anti-Patterns

| Anti-Pattern | Why It Fails | Instead |
|---|---|---|
| Recreating `ServiceBusClient` per message | Connection setup is expensive | Singleton pattern; register in DI |
| ReceiveAndDelete for critical workloads | Messages lost on consumer crash | Use PeekLock with manual complete |
| SQL filters everywhere | Each evaluation costs 1 credit; reduces throughput | Use correlation filters when possible |
| Ignoring DLQ messages | Processing failures go unnoticed | Monitor DeadletteredMessages metric; set up alerts |
| Standard tier for predictable latency | Shared infrastructure with throttling risk | Use Premium for production workloads |
| Large prefetch with slow consumers | Lock expires before processing; messages redelivered | Size prefetch to processable within lock timeout |

## Reference Files

- `references/architecture.md` -- Namespace, queues, topics, subscriptions, sessions, filters, transactions, scheduled messages, deferral, auto-forwarding, Premium tier, geo-DR, partitioning, AMQP 1.0
- `references/best-practices.md` -- Tier selection, connection management, receive modes, prefetch tuning, session design, security, SDK patterns (.NET, Python, Java, JS), Bicep/CLI deployment
- `references/diagnostics.md` -- Throttling, DLQ investigation, session issues, lock expiration, connectivity, Azure Monitor metrics, Log Analytics queries, cost analysis

## Cross-References

- The `overview` skill -- cross-broker comparisons and event-driven architecture design