Kaizen 22 is a debugging session, not a retrospective. Gerhard Lazu joins Jerod Santo and Adam Stacoviak to work through out-of-memory errors hitting changelog.com in production, inspect a new Pipedream instance status checker they built, and chase down an anomaly: one user in Asia repeatedly downloading a single episode at a scale that breaks normal patterns.
The episode references 'Stuff Goes Bad: Erlang in Anger', which signals the OOM conversation goes deeper than restarts and dashboards. The Erlang fault-tolerance model, 'let it crash', is the philosophical frame here, and hearing it applied to a real production system with real traffic data is where the value sits. The Pipedream checker analysis is a secondary thread worth following for anyone building lightweight status infrastructure without a dedicated observability stack.
This is episode 124 of Changelog Friends, the 22nd Kaizen installment. The full discussion is open on GitHub at changelog.com discussions thread 554. Read the show notes for the Abacus.ai and Mole references, both of which appear without much context in the episode and deserve independent investigation.
[READ ORIGINAL →]