Service Bus vs Storage Queues vs Event Grid vs Event Hubs

Verdict: Commands that must land exactly once go on Service Bus. Reactive triggers go on Event Grid. Streaming with replay goes on Event Hubs. Storage Queues only for queues over 80 GB or when an audit trail is required.

Criterion	Service Bus	Storage Queues	Event Grid	Event Hubs
Semantics	Commands; at-least-once delivery, at-most-once processing	Simple async jobs	Discrete events, push or pull	Streaming log, replay by offset
Max message	256 KB Standard; 100 MB Premium	64 KB	Event envelope, not payload transport	Event batches, throughput-unit metered
Ordering and dedup	Sessions (FIFO), duplicate detection, transactions, DLQ	None; at-least-once only	No ordering guarantee	Per-partition ordering
Retention or replay	TTL up to 14 days	Up to 7 days	24-hour push retry window	7 days Standard; 90 days Premium and Dedicated
Choose when	Order, payment, reservation flows	Queue may exceed 80 GB; audit trail needed	Azure resource change triggers; consumer behind Private Link uses pull	Kafka migration without code changes; telemetry backfill

Rules

Choose Service Bus for command workloads: peek-lock, duplicate detection, transactions and a built-in dead-letter queue come only there.
Choose Service Bus Premium when messages exceed 256 KB (up to 100 MB), or VNet isolation and geo-disaster recovery are required.
Choose Storage Queues above 80 GB or for audit trails; individual messages cap at 64 KB.
Choose Event Hubs to retire Kafka clusters: existing clients repoint to its Kafka endpoint without code changes.
Choose Event Grid pull delivery when the consumer cannot expose a public endpoint or must connect over Private Link.
Publish CloudEvents 1.0 for cross-vendor interoperability; the Event Grid schema is proprietary.

Traps

256 KB is the Service Bus Standard limit, not the Storage Queues limit - Storage Queues cap at 64 KB.
Event Hubs Standard caps replay at 7 days; Capture to Blob is not in-stream retention. A 30-day replay window needs Premium or Dedicated.