[@azure/functions-extensions-servicebus] Message settlement increases the latency of individual invocations compared to autocomplete

[rob-moos]

The gRPC calls to settle messages increases the latency of individual invocations (20ms+). This is much less performant for a high throughput function and means that we are less likely to adopt the SDK. Would the recommendation be to use autocomplete on high throughput functions and only utilise the methods for deadlettering and abandoning?.

Additionally the type seems to be incorrect for service bus topic triggers, where the sdkBinding property is not present on the ServiceBusTopicFunctionOptions type.

[swapnil-nagar]

rob-moos

Thank you for reaching out with your observation.

Manual settlement (complete, abandon, deadletter, etc.) requires an extra gRPC call from the Node.js worker to the Functions host, which explains most of the ~20 ms overhead. This is expected with the current design.

For high-throughput scenarios, recommend autoCompleteMessages: true. Use manual settlement only when the app needs specific control, such as immediate dead-lettering, modifying properties, or deferring messages.

For batches, run settlement calls in parallel with Promise.all() to reduce the overall overhead.

We will release a library update that addresses the other issue the customer raised(on ServiceBusTopicTriggerOptions).

[rob-moos]

Swapnil Nagar (@swapnil-nagar) Thank you for the help, I'll make sure to use autoCompleteMessages: true in our high throughput happy paths.

I've got a small feedback point RE the README: It might be worth further clarifying what the performance optimisations are, this is where some of the confusion comes from as a user of the extensions. My initial expectation without knowing the underlying design was that there would be performance improvements for all methods of message settlement, whereas message settlement latency actually increases with the main benefits provided being the control over settlement.