Ray Internals: A Peek at `ray.get`
summary
Deep dive into what ray.get actually does: precise lifecycle of an object, how small objects stay at the owner while large objects live in a distributed store, and how the distributed object directory locates copies; guarantees a primary copy pinned until the owner allows unpinning; shows a live demo that creates 20 ~100 MB arrays on a 1 GB object store, triggers OOM because of 17 local references kept by a list, and fixes it by dropping those Python refs so automatic ref counting frees memory.
AI-written summary · about
chapters (7)
- 0:00Introduction to Ray memory
- 1:03Architectural overview
- 2:33Small vs. large objects
- 4:35Lifecycle of an object
- 7:09Handling object operations
- 13:00Memory deallocation
- 14:46Debugging memory issues
related talks
24mRay Internals: Object Management with the Ownership Model
11mKeynote Q&A
26mData Processing on Ray
24mRay: A System for High-performance, Distributed ML Applications
37mEvolving Ray Core: Scalability, Reliability, and Compiled Graphs
15mRay: A Distributed Compute Engine for AI
28mA Deep Dive into Ray’s Scheduling Policy
29mRay:A Distributed Computing Platform for Reinforcement Learning