Kattie Gumanji

This talk is about the changes in OSS to start to expose APIs and change the way they interoperate.

Example:
- Docker Swarm
- Mesos
- K8s

K8s exposed APIs and created a whole ecosystem around it with lots of different tools popping up.

This is going to be very focussed on k8s, she is a CNCF board commitee member.

# Practical Past
 - CNI
  -- the story around here seems to be around good structure/apis and a working ecosystem --
   Calico
    has network policies enforcements
   flannel
   Cilium
    gaining momentum
   weaveNet

 - CRI
  This case was harder, the way the container execution was done was very intertwined into code and only a
  specific runtime was the one used at large.
  It was clear the need for a CRI, abstraction layer over the container runtime execution engine.
   - containerd
     industry standard, CNCF project
   - cri-o
     lightweight, roughly equivalent to containerd in features
   - gVisor
     from google, uses kernel syscall tracing to further isolate
   - firecracker
     from AWS, uses HW virtualization and creates effectivly micro-VMs with overhead of 7mb only.


# Innovation Wave
    SMI: service mesh interface
        normalization of this design pattern of having:
         infra layer to manage svc2svc traffic within a distributed environ.
        APIS:
            traffic policy
            traffic telemetry
            traffic management
        ISTIO
        LINKERD (incubating CNCF)
        CONSUL

    CSI: container storage interface
        plugable interface to consume external storage
        it is in GA since 1.13

        storageos
        rook (incubating to CNCF, also uses ceph)
        ceph
        OpenEBS

     ClusterAPI: Cluster Provisioning, cloud agnostic
        kubeadm
        kubespray
        kops
        tectonic
        (the image on the screen was: turtles on top of turtles)

        What if you want to migrate your cluster between infra providers ?
        Cross-cloud deployments ?
        Tool migration ?
        Challenges: in some regions: like China & Russia, for some reason are on the slides..

        the goal: declarative APIs to declarative create, configure and manage k8s clusters.
        the current api (in alpha: v1alpha) is supported by google cloud, aws, baidu-cloud, digital ocean, and some little others.

        this needs:
            cluster ap crds
            bootstrap providers (like kubeadm)
            infra providers  (aws, gke, etc..)
            .. requires a k8s cluster to create clusters.. lol (turtles all the way down)

            CIDRs for pods and services, DNS suffix
            Machines: kubelet and control plane version
            MachineSet: declare the Machine/Host resources, it is like a replicaSet or deamonSet
            MachineDeployment: declare the machine/host cconfigs I guess

# Emergence of Interfaces
    Extract core binaries and by creating an interface, expose them.
    Huge impact on vendors, users, community.
        - space for Innovation
        - extensibility, space for choice that best fits the user.
        - interoperability is important for community.

    CloudNative Interfaces has been key for the healthy growth of k8s.