Customizable Cluster Metrics Made Easy with Raydar and Anyscale
summary
Raydar is an open source pip package that you can pip install into a Ray runtime to capture task metadata outside the Ray GCS and tracks 250,000+ tasks live, exposing them as DataFrames or live Perspective tables. It ingests Prometheus node metrics, publishes custom user tables for PyTorch Lightning and Ray Train (tested on 40 GPU workers), helped find a data‑loading bottleneck, and integrates with Perspective and Anyscale with config export and persistence.
AI-written summary · about
chapters (8)
- 0:00Introduction and background
- 3:54What is radar and why built
- 7:23Live task tracking with Perspective
- 10:12Cluster metrics with Grafana alternative
- 13:03Backend architecture of task tracker
- 16:30Custom ML metrics with PyTorch Lightning
- 20:41When to use radar over other tools
- 23:08Future development and Q&A
related talks
21mHow Ray and Anyscale Make it Easy to do Massive-scale ML on Aerial Imagery
25mData Science at Scale: From Notebook to Production with Airflow and Ray
20mThe Quick Journey to Using Ray: How We Implement Ray and Anyscale to Speed up our ML Processes
26mRayDP: Build Large-scale End-to-end Data Analytics and AI Pipelines Using Spark and Ray
31mNext-Generation Data Science Workflows Using Ray
21mRay Train on Anyscale: Observability and Reliability at Scale
31mFrom OOMs to 2x Throughput: Scaling Multimodal Data Curation on Ray
26mData Processing on Ray