Performance questions

What throughput can I expect?

There is no workload-independent throughput figure on this page. Report the actual hardware, storage, versions/features, message sizes, queue layout, client mode, flush/replication policy, duration, errors, and completed consumption. A short producer burst is not sustainable end-to-end throughput.

How do I size or tune a deployment?

Capacity and performance provides storage and catch-up calculations, bottleneck observations, and a repeatable comparison method. Monitoring explains the signals used to locate the limiting stage.

Does increasing message size or consumer count improve throughput?

A larger maximum message size changes a limit; it is not a batching operation. Additional consumers help only when queue assignment, ordering, and downstream capacity permit more completed work. See sending APIs and consumer modes.

Can I switch flush mode to reduce latency?

The change affects durability. Compare storage semantics and HA requirements before treating a different acknowledgement contract as a performance result.