Thoughtworks Technology Podcast · Thoughtworks

Durable computing: What is it and why now?

·38 min·2 clips

John Coleman defines durable computing as a program's ability to recover its state and continue from where it left off.

1. The Thoughtworks Technology Podcast episode "Durable computing: What is it and why now?" covers platforms such as Temporal and Restate that recover and continue distributed processes after failures. 2. Host Alexei joins Brandon Cook, a Principal Software Engineer based in New York, and John Coleman, a lead consultant in Thoughtworks' Bangkok office, who each ran hands-on assessments or proofs of concept with these platforms. 3. Their core thesis is that the ideas are not new, since John notes ACID transactions and Akka actor replay, but platforms now bundle them so teams write business code and get the guarantees. 4. Brandon describes a client assessment where the event-driven architecture handled the happy path but not the "sad paths," which led to a platform team assessing these tools. 5. John explains that platforms differ in granularity: some recover a program's internal memory, while others record call results and restart the whole process, which affects replay latency. 6. The three discuss evaluation criteria, including SaaS versus self-hosting for sectors like financial systems, a single-binary deployment such as Restate versus Temporal, SDK language support, and fit with fan-out or long-running workflows. 7. They flag lock-in with AWS Step Functions, Azure's offering and AWS durable lambdas, and Brandon's client wanted durable lambdas because the organisation was already embedded in AWS. 8. Brandon stresses that platforms rely on determinism, so a replayed step could process a financial transaction twice unless the consumer side is idempotent, and he says teams often wrongly assume the producer's idempotency is enough. 9. They note that platforms have not solved versioning of months-long workflows or business-level failures, that debugging shifts from stack traces to event history, and that durable agents can tear down while waiting days for a human response. 10. Suited to engineers and architects weighing workflow platforms for distributed systems.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Thoughtworks Technology Podcast