https://www.youtube.com/watch?v=0KyLG3UXhdw
At 8:10 talks about how a platform group is created
At 14:10, talks about using telemetry to determine whether to go forward or rollback. Says they are not there yet.
At 18 minutes starts talking about what makes labels special.
talk-microservices-with-deis#control-loop1 At 21 talks about the concept of a control-loop. Where you attempt to get to the desired state over and over. talk-microservices-with-deis#control-loop1
The controller is the core primitive at the heart of self healing systems.
At 26 and he talks about daemon sense (sp?) where they can have things running on each of the whatever container instance is on Kubernetes
talk-microservices-with-deis#not-fan-of-running-databases-in-containers1 At 27:30 he says he is not a fan of running databases in containers. talk-microservices-with-deis#not-fan-of-running-databases-in-containers1
Namespaces for logical isolation. I love this idea.
At 28:30 putting nodes into maintenance mode, so all the work drains away from them.
At 31:50 and starts talking about deis features that are on top of Kubernetes
At 38 talks about how helm helps with non-12-factor apps, like statefull services which can have persistent volumes. Helps in setting up and managing those.
At 41:50, Deis itself runs in Kubernetes as privileged containers.
At 42:20, health checks are built into the Kubernetes platform. (TODO - what does this mean? What kind of health checks? Look into)
At 44:45, managing an append-only ledger to see who did what and when. (Are these commands? Are they derived off of audits?)
At 46 you can have tags in Deis and they allow you to ensure that a given container will only run on a particular type of host, for example
At 49:20 talked about local development. Does not suggest doing local development, and talks about 12-factor app development/production parity.
At 56:10, talks about how they prioritize feature development.
At 58:10 he talks about REST endpoints using TLS even within the cluster, and how some people use a message bus instead because it's much lower latency. Also, people are starting to consider GRPC.
At 1:04:15 talks about people building their own custom platforms, like how we have been doing at Instacart