Scaling Helyx: From Memory Bottlenecks to High-Performance Anomaly Detection From hitting memory walls to breaking resource barriers: The story of how we transformed our architecture and strengthened …

Trendyol Tech Blog offers a glimpse into the technology and engineering practices powering one of Turkey's leading e-commerce platforms. With a focus on scalability, performance, and innovation, Trendyol Tech shares insights into microservices architecture, data engineering, and machine learning. Developers can learn about building high-traffic e-commerce systems, optimizing delivery pipelines, and leveraging  technologies to drive business growth.

Trendyol Tech

Trendyol's Helyx anomaly detection service faced critical memory bottlenecks when user adoption tripled, causing 30% of training jobs to fail with OOMKilled errors. Horizontal scaling couldn't solve the problem because individual jobs were processing massive datasets that exceeded the 16GB node limits. The team partnered with their ML Platform team to integrate Valet, a system running on high-capacity nodes (up to 200GB memory), transforming their service from a heavy worker into a lightweight orchestrator. This architectural shift enabled job isolation, on-demand resource provisioning, and efficient handling of large-scale ML workloads.

Scaling Helyx: From Memory Bottlenecks to High-Performance Anomaly Detection

The Architectural Shift: From Worker to Orchestrator

We believe in the power of an inclusive workplace. Our platform is for everyone, and so is our workplace. Each and…