Meta’s large-scale serverless applications, like Facebook and Instagram, are built by many independent teams and countless engineers. This distributed development model accelerates product development but can also lead to chaos without sufficient management and governance around how team releases impact capacity consumption. In this talk we describe how we’ve adapted standard FinOps practices around budgets and cost controls to work at a granularity beyond tags and assets.
Large Launch Request Tool (LLRT) empowers teams to manage capacity at the granularity of a feature while abstracting the underlying cloud architecture and assets. LLRT surfaces insights into expected usage directly to engineers and product owners before the feature is widely released, similar to what feature experimentation does for product metrics. This fundamentally shifts how engineers and product management evaluate release decisions. Previously decisions focused solely on increasing product metrics, have now shifted to an ROI based model taking into account associated capacity costs.