The Stack Overflow Podcast Durable execution: autosave for your microservices
16 snips
May 20, 2025 Jeremy Edberg, CEO of DBOS and former tech leader at Reddit and Netflix, and Chen Li, Co-founder and researcher from Stanford and MIT, dive into the world of durable execution. They discuss how DBOS Transact simplifies error recovery in microservices, crucial for today’s demanding AI applications. The conversation highlights the significance of database checkpointing and the role of tools to enhance reliability in machine learning pipelines. Their insights illuminate the need for resilience in an increasingly complex software landscape.
AI Snips
Chapters
Transcript
Episode notes
Jeremy Edberg's Career Journey
- Jeremy Edberg shared his winding career from dropping out of college to key early roles at eBay, Reddit, Netflix, and AWS Lambda outside Amazon.
- He joined D-Boss after being impressed by the founders' work on durable execution research and technology.
Durable Execution Explained
- Durable execution means checkpointing app state to persist every step so nothing is lost or duplicated.
- It ensures reliability for sequences like e-commerce checkouts, preventing double charges and lost orders.
Prevent Duplicate Executions
- Use databases to store execution state and check if steps were previously executed to skip duplicates.
- Checkpoint outputs after each step to guarantee exactly-once execution through idempotency.