Workflows@Facebook: Powering Developer Productivity and Automation at Facebook Scale
summary
Facebook runs two complementary engines: a dependency layer called DAC that sits on Fox distributed priority queue with MySQL state and shard manager, and an async platform that submits DAGs. DAC enables 2 billion workflows a day while async workflows handle about 10 million workflows a day. Centralized CWS uses an immutable event stream and event pruning/printing for observability and long runs. Teams migrate FBJ→CWS, Enterprise→async, SV→DAC and plan a unified workflow system reusing Fox and DAC to reduce duplication and cover diverse authoring models.
AI-written summary · about
chapters (12)
- 0:00Workflows at scale
- 1:45Scale drives automation
- 3:30Workflow frameworks benefits
- 5:14Workflow system diversity
- 7:01Async workflows and DAC
- 11:34Workflow adoption success
- 14:43DAC service internals
- 16:51Workflow authoring styles
- 17:54Orchestration centralization
- 20:42Event stream optimizations
- 23:49Future of Facebook workflows
- 26:19Consolidation next steps
related talks
22mAsynchronous Computing at Scale: Overview and Learning
01h 03mDeveloping C++ @ Facebook Scale
27mSupercharging Backstage Scaffolder for Workflows
01h 01mPanel: Facebook's Distributed Database Evolution
28mHow Did Backstage Make 500+ Developers Happy in a Highly Regulated Enterprise
35mC++ Workflow: Task-based Parallel Computing and Asynchronous Networking Engine
52mEvent Driven Workflows
30mAfter the Click