Kranti Parisa

Krishna Vishal

@krishvishal Editor

Building Apache Iggy: How Rust Powers Ultra-Low Tail Latencies with io_uring

Submitted Feb 20, 2026

Apache Iggy is a next-generation message streaming engine built entirely in Rust, delivering stable sub-millisecond P99 latencies with high throughput on a single node. In this talk we’ll show what it takes to get there and how Rust’s ownership model shaped every architectural choice.

We’ll start with Iggy’s core model: an append-only log, partitions as the unit of parallelism, and zero-copy serialization that doubled consumer throughput to 4 GB/s. Then we’ll dive into the performance work that makes Iggy different: why epoll and Tokio’s work-stealing scheduler can’t deliver predictable latencies for block device I/O, how we migrated to io_uring with the compio runtime, and the Rust-specific challenges we hit - RefCell panics across .await boundaries that forced us to redesign state management, and the shared-something hybrid architecture we landed on using left-right crate. We’ll show benchmarks: 92% P9999 latency improvement, consistent 1 GB/s+ throughput across partition counts.

We’ll wrap up with a look at what’s next: clustering via Viewstamped Replication (VSR) and the deterministic simulation harness we are building to test it.

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

A community of Rust language contributors and end-users from Bangalore. We have presence on the following telegram channels https://t.me/RustIndia https://t.me/fpncr LinkedIn: https://www.linkedin.com/company/rust-india/ more