How to Build a Reliable Kafka Data Processing Pipeline, Focusing on Contention, Uptime and Latency

Shifting workloads from synchronous to asynchronous can simplify the operational cost of high-throughput HTTP services. But understanding the evolution of performance metrics in the world of complex, high-concurrency, asynchronous distributed systems can be quite challenging.

In this talk, I'll tell you how OneSignal improved the performance and maintainability of its highest-throughput HTTP endpoints (backed by a Kafka consumer in Rust) by making it an asynchronous system. We will cover:

  • How metrics changed when the system went from sync to async
  • Unique sharding strategies to maximize concurrency and performance, while maintaining consistency for Kafka consumers
  • System-level constraints from Postgres infrastructure determining the Kafka scaling strategy.


Lily Mara

Engineering Manager @OneSignal

Lily Mara is an Engineering Manager at OneSignal in San Mateo, CA. She manages the Infrastructure Services team, which is responsible for in-house services used by other OneSignal engineering teams. Previously she was a software engineer at OneSignal, leading the efforts to create OneSignal's integration with Mixpanel, develop the outcomes system, and improving performance and code simplicity through refactoring efforts. Lily also worked as a software developer at Kroger, working on Kroger’s online grocery ordering system as well as internal development tools to aid other teams in deployments, monitoring, and local development environments.

Lily is the author of Refactoring to Rust, an early-access book by Manning Publications about improving the performance of existing software systems through the gradual addition of Rust code.

Read more
Find Lily Mara at:


Wednesday Jun 14 / 10:35AM EDT ( 50 minutes )


Salon A-C


Kafka Reliability Architecture Performance


From the same track

Session Architecture

Reliable Architectures Through Observability

Wednesday Jun 14 / 02:55PM EDT

We want our systems to be reliable, but testing alone isn't enough. In a complex, multi-service system, it's impossible to test your way to correctness. That's why we need observability. Observability is the ability to see what our code is doing, in production and in development.

Speaker image - Kent Quirk
Kent Quirk

Staff Engineer

Session Architecture

Building an Architecture to Predict Customer Behavior in a Revenue-Critical System

Wednesday Jun 14 / 01:40PM EDT

At Neon digital bank in Brazil, we strive to make revenue-impacting predictions based on customer behavior. Building a low latency and high availability distributed system that meets this requirement becomes especially challenging.

Speaker image - Yves Junqueira
Yves Junqueira

Distinguished Software Engineer @Neon

Session Developer Environment

Architecting a Production Development Environment for Reliability

Wednesday Jun 14 / 04:10PM EDT

At Meta, developers use a combination of development servers, including virtual machines and physical hosts, as well as on-demand containers to perform their daily software engineering work.

Speaker image - Henrique Andrade
Henrique Andrade

Production Engineer @Meta

Session Cloud Architecture

Survival Strategies for the Noisy Neighbor Apocalypse

Wednesday Jun 14 / 05:25PM EDT

Noisy neighbor issues are a common challenge for multi-tenant platforms, leading to resource contention, performance degradation, and costly downtime for other tenants sharing the same resources.

Speaker image - Meenakshi Jindal
Meenakshi Jindal

Staff Software Engineer @Netflix


Unconference: Designing Modern Reliable Architectures

Wednesday Jun 14 / 11:50AM EDT

What is an unconference? An unconference is a participant-driven meeting. Attendees come together, bringing their challenges and relying on the experience and know-how of their peers for solutions.