Keynote: DurableServer: Always Running Somewhere
summary
DurableServer extends GenServer with automatic state persistence, cluster-aware placement and load balancing, using object storage (S3 / Tigris) or a local EKV backend with CAS Paxos compare-and-swap. The open source stack runs millions of sprites (over 1M buckets), demos 500,000 simulated riders with 50,000 drivers, survives a node bounce that drops ~12,000 drivers and restarts them, and supports on‑demand SQLite + Lightstream DBs for hot clones and sticky placement.
AI-written summary · about
chapters (20)
- 0:00DurableServer intro
- 1:27What Sprites are
- 3:52Multi-region necessity
- 8:08The messy birth of DurableServer
- 10:18Durable objects in Elixir
- 13:29Load balancing need
- 15:37DurableServer as OTP lib
- 18:49Metadata and process discovery
- 21:16Live demo
- 27:43Re-homing processes
- 30:12Always running meaning
- 33:21Routing and bottlenecks
- 35:27Storage cost
- 37:14EKV for cheaper storage
- 40:21CAS Paxos simplified
- 41:47Uber ride-sharing demo
- 45:39Durable state for ride-sharing
- 46:23Surviving failures with EKV
- 48:50Replacing Phoenix Presence
- 50:54Future of agents
related talks
45mThe simplest way to build resilient applications
35mBuilding resilient applications with Durable Execution
28mStateful Applications in the Age of Durable Execution
25mCloudstate—Towards Stateful Serverless
33mCloudstate Towards Stateful Serverless
48mThe simplest way to build resilient applications
01h 23mBuilding Highly-Scalable Spring Applications with In-Memory, Distributed Data Grids
44mThe Simplest Way to Build Resilient Applications