Leios proto-devnet adopts tx-firehose for throughput testing

What changed

The Ouroboros Leios proto-devnet has switched its throughput load generator from tx-centrifuge to tx-firehose. The new tool is described as a single-node, push-based generator that connects over node-to-client, finds its own UTxO through an address query, and exposes whether transaction submissions are accepted or rejected.

The load is then propagated to other devnet nodes through N2N transaction submission. A 30-second restart backoff was also added after a diagnosed mempool-descendant crashloop. The related dashboard work focuses on real serialized transaction size and an accepted-versus-rejected view.

Why it matters

For a prototype intended to test throughput, the quality of the load harness affects what the team can observe. A generator that can discover UTxO and surface submission outcomes gives operators a clearer view of whether a run is progressing or encountering rejects, and allows runs to restart on demand.

This is therefore a progress signal for Leios testing rather than a performance result. The record frames the work as tooling and observability needed to measure the scaling gains Leios is intended to deliver, but it does not establish those gains.

What to watch next

The most useful receipt will be the tx-firehose accepted-and-rejected dashboards, alongside any future reported throughput benchmarks from the proto-devnet. Those signals may show how the new harness behaves under load, but dashboard activity alone should be treated as instrumentation rather than evidence of a completed scaling outcome.