In this blog post, we will share our analysis of Spark Dependency Management at LinkedIn, highlight interesting findings, and present our design choice of using a simple user-level cache over more complex alternatives. We will also discuss our rollout experience and lessons learned. Finally, we will demonstrate the impact of accelerating all Spark applications at LinkedIn at the cluster level. Faster Spark jobs translate to increased data freshness, leading to an enhanced member experience by way of more relevant recommendations, timely insights, effective abuse protection, and other similar improvements.
If you work with Spark to any serious extent, you’ll want to read this post.