Waterpark: Distributed Actors vs the Pandemic
summary
A concrete build story: a custom platform called Waterpark runs on ErlangVM and Elixir across four data centers for HCA’s 185 hospitals and ~2,000 clinics, models a patient actor per person that holds weeks of HL7 events, and adds bloom filters, rendezvous hashing and process-pair replicas. The claim is continuous availability—continuously available in three weeks—hot code loading in production, zero planned or unplanned downtime for 18 months, and an open source HL7 library that shipped and worked.
AI-written summary · about
chapters (14)
- 0:00Introduction to Waterpark
- 3:01System architecture overview
- 4:34The Erlang VM advantage
- 6:28Actor model and scheduling
- 11:00Modeling patient data
- 13:58HL7 and bloom filters
- 18:03High availability strategies
- 22:17Cluster topology and hashing
- 24:28Distributed consistency
- 28:16Mailroom and state management
- 30:08Deployments and hot loading
- 32:24Universal server implementation
- 34:20Pandemic response and conclusion
- 36:57Q&A and closing thoughts
related talks
43mThe River
44mMedical Messaging: An Elixir For Healthcare
39mAn adventure in distributed programming
01h 03mThinking Erlang in a connected world
44mIdioms for Building Fault-tolerant Applications with Elixir
31mAn Open Erlang PaaS Done Right
53mIdioms for Building Distributed Fault-tolerant Applications with Elixir
44mThe NeverEnding Application