The CMS offline software and computing system has successfully met the challenge of LHC Run 2. In this presentation, we will discuss how the entire system was improved in anticipation of increased trigger output rate, increased rate of pileup interactions and the evolution of computing technology. The primary goals behind these changes was to increase the flexibility of computing facilities where ever possible, as to increase our operational efficiency, and to decrease the computing resources needed to accomplish the primary offline computing workflows. These changes have resulted in a new approach to distributed computing in CMS for Run 2 and for the future as the LHC luminosity should continue to increase. We will discuss changes and plans to our data federation, which was one of the key changes towards a more flexible computing model for Run 2. Our software framework and algorithms also underwent significant changes. We will summarize the our experience with a new multi-threaded framework as deployed on our prompt reconstruction farm for 2015 and across the CMS WLCG Tier-1 facilities. We will discuss our experience with a analysis data format which is ten times smaller than our primary Run 1 format. This "miniAOD" format has proven to be easier to analyze while be extremely flexible for analysts. Finally, we describe improvements to our workflow management system that have resulted in increased automation and reliability for all facets of CMS production and user analysis operations.