CRGC: Fault-Recovering Actor Garbage Collection in Pekko
summary
CRGC implements the first fault-recovering cyclic actor GC for distributed actor systems and ships a Java prototype in Apache Pekko. It records local actor updates, merges them into per-node shadow graphs, exchanges small undo logs on node failures, and recovers lost state by querying neighbors; it requires only FIFO delivery to the local GC. The design is formalized in TLA+ and proved soundness and completeness, and benchmarks show overhead competitive with weighted reference counting.
AI-written summary · about
chapters (8)
- 0:00Actors and challenges
- 3:51Actor GC uniqueness
- 7:40Detecting crashed nodes
- 9:47GC consistency challenges
- 12:35Recovering from crashes
- 15:24Shadow graph and undo logs
- 17:08Future work and improvements
- 18:33Q&A
related talks
18mReliable Actors with Retry Orchestration
55m“A lock-free concurrency toolkit for deferred reclamation and optimistic speculation"
26mAn Exceptional Actor System (Functional Pearl)
19mPlatinum: A CPU-Efficient Concurrent Garbage Collector
45mPony for Safe, Fast, concurrent programs
01h 03mErlang & Akka Actors - A story of tradeoffs
58m“The Landscape and Exciting New Future of Safe Reclamation for High Performance”
25mSCALAZ 8 VS AKKA ACTORS