Welcome to our eighth 'East Bay Cloud Native' meetup!
6:00: Grab a seat. Have some food and drinks and network.
6:30: Talk 1 - TensorFlow on Kubernetes (Dharmesh Kakadia)
7:30: Talk 2 - Extending K8s With CRDs and Operators (Chris Hein)
8:30: More networking until 9:00pm.
Talk 1 Description: I will briefly highlight some of our machine learning use cases, before diving deep into how we run Tensorflow based machine learning workflows on Kubernetes. It includes how we make cloud storage accessible for model training, how we store models currently and operational aspects of machine learning productization.
I will also demo our dev workflow for Tensorflow with an example and how we deploy prod pipeline through CI/CD. Finally I will touch upon current state of GPU support on k8s, building vs buying a machine learning platform and some of the learning we had along the way.
Speaker Bio: Dharmesh is a Software Engineer/Applied Scientist at Microsoft working on AI/Analytics platform. He has worked on various open source projects in cloud and data systems. Among other things, he is also author of Apache Mesos Essentials. He loves reading, hiking and above all deep conversations.
Talk 2 Description: Within the Kubernetes community, there has always been a desire to continue to add new features and functionality into the apiserver; you see this every day from the version of each primitive you use. This desire grew so much that the community introduced custom aggregate apiservers these allow you to build custom apis and run them on the master with coordination from the exposed apiserver, but they require you to run your own storage layer or convince your cluster admin to let you use the primary etcd cluster. The community then created ThirdPartyResources; these gave you the ability to create custom resources that were registered with the apiserver and were able to be run what CoreOS coined as the Operator pattern. TPRs, while great, eventually were rebuilt and moved into apiextensions and called CRDs. In this talk, we will build an Operator using CRDs, the k8s code-generation libraries and informer libraries which are generated.
CRDs and Operators are excellent for anything from managing some custom resource like public keys in an SSH authorized_key file or to help you to create easily deployable custom-off-the-shelf software that any cluster administrator or developer could deploy into their environment without having to conform to a predefined spec.
Speaker Bio: Chris Hein is a Sr. Developer Advocate for Kubernetes/EKS at Amazon Web Services. Before Amazon, Chris worked for a number of large and small companies like GoPro, Sproutling, & Mattel.
Please RSVP so the food can be ordered accordingly!
Software Development Director at Oracle Cloud