Gerhard Lazu returns for Kaizen 22, the 22nd iteration of Changelog's continuous improvement series, and this episode has a concrete problem at its center: out-of-memory errors hitting changelog.com in production. Lazu, Santo, and Stacoviak dissect the crashes, walk through a new Pipedream-based instance status checker built to catch failures earlier, and confront a genuine mystery: one user in Asia repeatedly downloading a single episode at a volume that breaks expected traffic patterns.
The OOM investigation is the core reason to listen. The team references 'Stuff Goes Bad: Erlang in Anger' as a diagnostic framework, which signals this is not a surface-level conversation. The Pipedream status checker discussion gets specific about how the tool was wired into their existing infrastructure, and the Asia download anomaly raises real questions about CDN behavior, bots, and how podcast metrics can quietly lie to you.
Kaizen episodes are rare in podcasting: a technical team doing honest post-mortems on their own production system in public, with the actual codebase open on GitHub at thechangelog/changelog.com. Discussion thread is at issue 554. If you care about running real infrastructure, not theorizing about it, start there.
[READ ORIGINAL →]