TransportTransactionMode levels are supported
TransportTransactionMode.level is only supported in Versions 7 and above of the Azure Service Bus transport. It requires destination and receive queues to be in the same namespace. Also, the transport type must be set to
SendsAtomicWithReceive guarantee is achieved by setting the
ViaEntityPath property on outbound message senders. Its value is set to the receiving queue.
ViaEntityPath is not empty, then messages will be added to the receive queue. The messages will be forwarded to their destinations (inside the broker) only when the complete operation is called on the received brokered message. The message won't be forwarded if the lock duration limit is exceeded (30 seconds by default) or if the message is explicitly abandoned.
ReceiveOnly guarantee is based on the Azure Service Bus Peek-Lock mode.
The message is not removed from the queue directly after receive, but it's hidden for, by default, 30 seconds. This prevents other instances from receiving the message. If the receiver fails to process the message within that time frame or explicitly abandons the message, then the message will become visible again. Other instances will be able to pick it up (effectively that rolls back the incoming message).
When transactions are disabled in NServiceBus then the transport uses the Azure Service Bus 'ReceiveAndDelete' mode.
The message is deleted from the queue directly after the receive operation completes, before it is processed, meaning that it's not possible to retry that message in case of processing failures. As transient exceptions occur regularly when integrating with online services, disabling retries when in unreliable mode is not recommended. This mode should only be used in very specific situations, when message loss is acceptable.
Using the supported transaction levels, the transport only provides an
at-least-once delivery model. Azure Service Bus provides a duplicate detection feature which will ensure the message is received exactly once if enabled. However, Duplicate Detection comes at the expense of throughput and is time constrained.