First Talk - Bayesian is all you need to monitor your third parties

Speaker: Clement Tiennot

Bayesian methods allow us to estimate statistical quantities while modeling the underlying uncertainty about our estimates. In this talk we showcase how to leverage such methods to improve our modeling of third party API error rates and detect when those are experiencing incidents. We describe a Bayesian Hierarchical model at a high level and how we apply it to our use case. No real prerequisite knowledge should be needed, but a background in data science or statistics might help understanding subtleties of the talk.

Second Talk - Accelerating Troubleshooting in Distributed Systems with Watchdog Outliers

Speaker: Kanvaly Fadiga

In the complex landscape of modern distributed applications, monitoring hundreds of services and infrastructure components presents a significant challenge. Even with robust monitoring and alerts, the infrastructure and applications are likely to undergo changes over time, potentially complicating the reliable detection of irregular behavior.
To address these challenges, the Datadog Data Science Team has developed Watchdog Outliers. This tool accelerates troubleshooting by quickly analyzing events (such as Logs, APM Traces, etc.) from user queries, pinpointing tags that may contribute to issues—whether it's a specific host, datacenter, or user. You will learn about common challenges and discover how to swiftly identify abnormal behavior, thereby reducing time to resolution.