Going Multi-Node
summary
Turns ExVenture into a cluster using LibCluster for node discovery, a Raft-based squabble leader election to pick a single leader that starts stateful zone processes, and PG for cross-node fan‑out; implements circuit breakers and “most‑once” handling for gossip (gossip das) and reduces session payloads. Three-node clusters rebalance zones on failure, scale from ~230 to ~3,500 concurrent bots after optimizations, hit ~50GB RAM limits on a 64GB desktop, and survive node crashes while reassigning sessions.
AI-written summary · about
chapters (14)
- 0:00Introduction to multi-node
- 1:06What is a MUD
- 7:37Single-node limitations
- 9:02Connecting nodes together
- 10:26Leader selection with Raft
- 12:16Squabble library explained
- 13:20Switching to global registry
- 14:23Spanning the cluster
- 15:48Problems encountered
- 17:56Performance bottlenecks
- 21:26Scaling to 3500 players
- 23:58Live clustering demo
- 27:13Wrap-up and resources
- 27:58Q&A session
related talks
39mAn adventure in distributed programming
44mCluster Consensus: When Aeron Met Raft
33mSelf-healing Clusters: Game of Nodes and the Scaling Throne
41mDistributed Systems explained with NodeJS
40mGoing Multi-Node with ExUnit
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data
25mSupporting 3 Availability Zones Stretch Cluster
21mThree Nodes and One Cluster