### Scaling Clusters Declaratively with Containers & Kubernetes #### Speaker: Brian Dorsey Linux containers (Docker, etc) are a great way to deploy Python processes and make sure all of their required dependencies are available. Once you start running a lot of containers, managing them becomes a challenge. Kubernetes (http://kubernetes.io) is an open source cluster manager and scheduler that simplifies deploying and managing the containers that make up your application. You declare the desired state, and Kubernetes does the work to keep it that way. It assigns resources, recovers from failures, and scales easily. Brian will give a brief overview of containers, introduce Kubernetes, and do a live demonstration of moving from a single Docker container to many containers running across a cluster of machines with Kubernetes. -------------------- ### Better Collaborative Data Science with CoLaboratory #### Speaker: Jeff Snyder CoLaboratory is an experimental open source project to which Google is a contributor. CoLaboratory adds collaborative editing of documents using Google Drive's realtime API, and user interface changes designed to facilitate collaboration, such as forms that allow users to create simple interfaces for other users to interact with code. Within Google, we continue to actively develop our internal fork of coLaboratory, which includes integration with proprietary Google technologies, to make life simpler for Google engineers. I will discuss how coLaboratory is used at Google, and the lessons we've learnt about how users collaborate around code and data. In the open source world, we have moved from developing coLaboratory, to upstreaming Google Drive integration into IPython (and the new Jupyter project). We are currently working on adding real time collaboration as an optional part of Jupyter. -------------------- ### Remote Procedure Calls from Monitoring Pipelines #### Speaker: Alex Perry Monitoring of production servers needs to be efficient enough not to impact the actual users of the system, yet deliver enough insight that any problems can be fixed efficiently. These often conflict, but Python's decorators can be used to deliver additional insight on demand with a tiny code impact. The demand need not be made until monitoring has had time to infer that the situation is undesirable. --------------------