1 · Why Decouple with Queues

Direct synchronous calls between services create tight coupling, cascading failures, and uneven load. Queues solve three fundamental problems:

• Temporal decoupling — Producer and consumer don't need to be online simultaneously.
• Load leveling — Queues absorb traffic spikes; consumers process at their own pace.
• Competing consumers — Multiple workers pull from one queue, scaling throughput horizontally.

2 · Service Bus vs Storage Queues

Rule of thumb: If you need topics, sessions, DLQ, or transactions → Service Bus. If you need cheap, massive backlogs → Storage Queues.

3 · Service Bus Concepts

Queue (Point-to-Point)

Each message is delivered to exactly one competing consumer. Ideal for work distribution.

Topic / Subscriptions (Pub/Sub)

Each subscription gets its own copy of the message. SQL-like filters on subscriptions reduce noise.

Dead-Letter Queue (DLQ)

Every queue/subscription has a DLQ sub-queue. Messages land there after max delivery attempts or TTL expiry. Monitor DLQ depth as a health signal.

4 · Message Patterns

• Point-to-point — Single queue, competing consumers. Order processing, job dispatch.
• Publish/Subscribe — Topic + subscriptions. Event fan-out (order placed → billing, inventory, notification).
• Request-Reply — Sender sets ReplyTo + CorrelationId; responder sends to reply queue.
• Priority queues — Separate queues per priority; consumers poll high-priority first.
• Claim-check — Large payload in Blob Storage; message carries the blob URI.

5 · Reliability Features

• Duplicate detection — Deduplicates by MessageId within a configurable time window (up to 7 days).
• Message deferral — Postpone processing; retrieve later by sequence number when ready.
• Scheduled delivery — Enqueue now, visible to consumers at a future UTC time.
• Transactions — Atomic send + complete across entities in the same namespace (via ServiceBusTransactionGroup).
• Auto-forwarding — Chain queues/subscriptions to build processing pipelines.
• Peek-lock vs Receive-and-Delete — Peek-lock gives at-least-once safety; Receive-and-Delete gives at-most-once speed.

6 · Service Bus Tiers

Choose Premium when you need network isolation (Private Link / VNet), predictable latency (no noisy neighbours), large messages, or Geo-DR pairing.

7 · Ordering & Sessions

Without sessions, Service Bus provides best-effort ordering. Enable sessions for guaranteed FIFO per session ID:

• Set SessionId on each message (e.g., order ID).
• Consumer accepts a session — gets exclusive lock on all messages in that session.
• Multiple sessions can be processed in parallel by different consumers while preserving per-session order.

Partitioned queues spread load across message brokers for higher throughput, but ordering is only within a partition (use sessions to pin related messages).