Rethinking How We Test Our Async Architecture
summary
CrowdStrike proves live testing of massive async pipelines by running real production data (over a trillion messages/day) through duplicate services and using no-op services (context flag skips I/O) plus API injection to replay messages synchronously. Each step records a memoir on context (thread-safe channel, custom JSON marshaller) so callers see exactly which writes would occur to Kafka, Cassandra, Redis, email, etc. The approach runs in production as a canary or blue/green check and yields concrete failures or confirmations, not mocks.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:21I'M TOM
- 1:15ASYNCHRONOUS ARCHITECTURES
- 5:25INTEGRATION TESTS Considering a more complex flow
- 9:38MY RULES FOR CONTEXT
- 14:33HOW DO WE NOOP?
- 18:50HOLD UP, I HAVE QUESTIONS
- 20:24IT MATTERS WHERE YOU PUT IT
- 27:13WHAT MAKES A MEMOIR
- 32:52TESTING WITH MEMOIRS
related talks
01h 02mThe Art & Joy of Testing in Production
28mHow Not-To Test in Production: Learnings from DoorDash
25mTesting in production: Ideas, experiences, limits, roadblocks
30mElevating Event-Driven Architecture: Boost your delivery with AsyncAPI and Microcks
30mHow to Test Asynchronous Code
48mTesting The Cloud
27mEffectively Testing Services
56mTesting at scale