A Node.js service faced performance issues due to improper handling of worker threads, which caused high resource consumption and server instability within a Kubernetes environment. By spawning multiple workers per CPU core instead of per allocated resource, and aggressively restarting them on errors, a positive feedback loop

12m read timeFrom engineering.zalando.com
Post cover image
Table of contents
A disrupted gaming nightNot so fast TarnishedDigging deeperBuilding better observability
3 Comments

Sort: