SREcon24 Europe/Middle East/Africa

Laura Nolan; Niall Murphy, Stanza


Effective load management is a core aspect of the SRE role. In this workshop, participants will be introduced to a number of Envoy proxy features that are used for loadshedding and isolation, such as circuit breaking, adaptive concurrency, and ratelimiting. Participants will also use custom Go plugins to perform loadshedding. As part of the practical element of the workshop, participants will interact with Envoy configurations and status/control pages and endpoints, as well as Envoy’s telemetry.


https://www.usenix.org/conference/srecon24emea/presentation/nolan

Leila Vayghan, Shopify


This talk is about enhancing the search infrastructure of Shopify, a large-scale ecommerce platform that supports over 3 million merchants and handles more than two petabytes of data.

This talk explains how we leverage Kubernetes on Google Cloud Platform to ensure high availability and performance, crucial for maintaining our platform's robust search functionality. It will also elaborate on our innovative approach using dedicated reindexing nodes within existing clusters, which significantly improves indexing and reindex performance while cutting infrastructure costs. We will explore the application of Kubernetes Event-Driven Autoscaling (KEDA) to dynamically manage resource allocation, enhancing operational efficiency and reducing on-call fatigue. This strategy not only supports seamless user experiences but also boosts Gross Merchandise Value (GMV) and revenue through improved system responsiveness.

This presentation is ideal for those involved in managing large-scale data systems or interested in advanced Elasticsearch optimizations.


https://www.usenix.org/conference/srecon24emea/presentation/vayghan

Moeid Heidari


This research tackles the limitations of traditional autoscaling systems, which typically operate within a single cloud provider. We propose a new Kubernetes autoscaling operator that dynamically adjusts resources across multiple cloud platforms and on-premise systems. By integrating with various provisioning systems and allowing user-defined scaling strategies, this operator addresses the inefficiencies and vendor lock-in issues of conventional solutions. Our approach not only enhances scalability and system resilience but also improves cost-efficiency, as demonstrated by a significant increase in system availability. Metrics are collected and analyzed to predict scaling needs, ensuring optimal performance and resource utilization.


https://www.usenix.org/conference/srecon24emea/presentation/heidari