Using distributed trace data to solve performance and operational challenges

Description

Like most modern large-scale applications, Pinterest is built on a microservices architecture. In this scheme, a number of services work together to serve a single user request. Debugging performance and architectural problems in this environment can be challenging. Distributed tracing has emerged as the indispensable tool and solution to address these challenges.

Pinterest recently deployed Pintrace, a Zipkin-based distributed tracing system, to record end-to-end performance data across the execution path of requests, from mobile applications to backend services. Pintrace has evolved over time as its users find new data and as new subsystems integrate with the company’s tracing systems. Pinterest has also built tools for visualization, feature extraction, aggregation, and analysis of trace data, which help enable use cases that wouldn’t have been possible with traditional tooling, such as root-cause analysis, latency analysis, and regression analysis.

Naoman Abbas offers an overview of tools Pinterest built to process trace data and the use cases they’ve enabled and shares some real-world examples. Join in to learn how to apply these techniques to your own challenges.