Strange Loop Conference | "Hodor: Detecting and Addressing Overload in LinkedIn Microservices" by Bryan Barkley @StrangeLoopConf | Uploaded October 2022 | Updated October 2024, 1 week ago.
When pushed hard enough any system will eventually suffer, and ultimately fail unless relief is provided in some form. At LinkedIn, we have developed a framework for our microservices to help with these issues: Hodor (Holistic Overload Detection & Overload Remediation). As the name suggests, it is designed to detect service overloads from multiple potential root causes, and to automatically improve the situation by dropping just enough traffic to allow the service to recover. Hodor then maintains an optimal traffic level to prevent the service from reentering overload. All of this is done without manual tuning or specifying thresholds. In this talk, we will introduce Hodor, provide an overview of the framework, describe how it detects overloads, and how requests are dropped to provide relief.
Bryan Barkley
Sr. Staff Engineer at LinkedIn
Bryan is an engineer at LinkedIn working with service infrastructure teams to improve the resiliency and availability of systems. He has been working with server side Java for over two decades. Prior to LinkedIn he was an engineer at TripAdvisor, Time Inc., Monster, and a variety of smaller startups.
------ Sponsored by: ------
Stream is the # 1 Chat API for custom messaging apps. Activate your free 30-day trial to explore Stream Chat. gstrm.io/tsl
When pushed hard enough any system will eventually suffer, and ultimately fail unless relief is provided in some form. At LinkedIn, we have developed a framework for our microservices to help with these issues: Hodor (Holistic Overload Detection & Overload Remediation). As the name suggests, it is designed to detect service overloads from multiple potential root causes, and to automatically improve the situation by dropping just enough traffic to allow the service to recover. Hodor then maintains an optimal traffic level to prevent the service from reentering overload. All of this is done without manual tuning or specifying thresholds. In this talk, we will introduce Hodor, provide an overview of the framework, describe how it detects overloads, and how requests are dropped to provide relief.
Bryan Barkley
Sr. Staff Engineer at LinkedIn
Bryan is an engineer at LinkedIn working with service infrastructure teams to improve the resiliency and availability of systems. He has been working with server side Java for over two decades. Prior to LinkedIn he was an engineer at TripAdvisor, Time Inc., Monster, and a variety of smaller startups.
------ Sponsored by: ------
Stream is the # 1 Chat API for custom messaging apps. Activate your free 30-day trial to explore Stream Chat. gstrm.io/tsl
