Uploaded November 2025 | Updated November 2026, 19 hours ago
Slides: drive.google.com/file/d/1TTWdAHsoK5fgtMVpdYg6keyn_lmqN3kd/view?usp=sharing
At Ray Summit 2025, Janet Li and Simran Mhatre from Anyscale share the latest advancements in observability for Ray—designed to help developers debug, optimize, and understand distributed AI workloads with far greater clarity.
They begin by explaining why contextual observability is essential. As Ray becomes the standard framework for distributed AI applications, users increasingly encounter the inherent complexity of large, multi-node systems. Purpose-built observability tools are crucial for diagnosing issues such as resource bottlenecks, task failures, and memory pressure.
Janet and Simran then introduce major new improvements to observability on Anyscale, including scalable and persistent dashboard views for Ray Core, Ray Train, and Ray Data. They walk through the architecture powering these dashboards and show how they operate securely while keeping all data within the customer’s cloud environment.
The speakers also unveil the open-source Ray Export API, which allows users to persist, analyze, and integrate the same events shown in the Ray dashboard into their own monitoring and analytics systems.
A live demo demonstrates how these tools can be used to debug real-world issues—from out-of-memory errors to inefficient resource utilization—making Ray workloads more transparent, reliable, and easier to optimize than ever before.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
Slides: drive.google.com/file/d/1TTWdAHsoK5fgtMVpdYg6keyn_lmqN3kd/view?usp=sharing
At Ray Summit 2025, Janet Li and Simran Mhatre from Anyscale share the latest advancements in observability for Ray—designed to help developers debug, optimize, and understand distributed AI workloads with far greater clarity.
They begin by explaining why contextual observability is essential. As Ray becomes the standard framework for distributed AI applications, users increasingly encounter the inherent complexity of large, multi-node systems. Purpose-built observability tools are crucial for diagnosing issues such as resource bottlenecks, task failures, and memory pressure.
Janet and Simran then introduce major new improvements to observability on Anyscale, including scalable and persistent dashboard views for Ray Core, Ray Train, and Ray Data. They walk through the architecture powering these dashboards and show how they operate securely while keeping all data within the customer’s cloud environment.
The speakers also unveil the open-source Ray Export API, which allows users to persist, analyze, and integrate the same events shown in the Ray dashboard into their own monitoring and analytics systems.
A live demo demonstrates how these tools can be used to debug real-world issues—from out-of-memory errors to inefficient resource utilization—making Ray workloads more transparent, reliable, and easier to optimize than ever before.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com










