SYSTEM Cited by 1 source
AWS SageMaker Studio¶
SageMaker Studio is AWS's web-based IDE for ML development, part of the SageMaker AI product line. It hosts JupyterLab notebooks, code editors, and experiment tracking on managed compute, with per-user "Spaces" and Domain-level networking isolation.
Studio Domains expose configurable networking (VPC mode, route targets, subnets) but by default restrict inbound connections to the Domain's network interfaces, which can break workloads that expect bidirectional connectivity between the Studio notebook and external clusters.
Lyft / LyftLearn cross-cluster Spark¶
On Lyft's LyftLearn 2.0 migration, the Spark driver in a SageMaker Studio notebook had to talk bidirectionally with Spark executors running on a separate EKS cluster. Default Studio networking blocked the inbound executor→driver leg; AWS partnered with Lyft to introduce networking changes to the Studio Domains in Lyft's account so the required inbound EKS→Studio traffic was allowed. Result: Spark performance and interactive UX unchanged from the pre-migration same-cluster setup (Source: sources/2025-11-18-lyft-lyftlearn-evolution-rethinking-ml-platform-architecture). This is a concrete instance of cross-cluster networking where the managed service required provider-side intervention.