20170329_kubeCon_1.2.3
20170329_kubeCon_1.2.3
2 - Presentations
2.3 - Hosting and managing K8s clusters @Google
GKE
Hosted and Managed vanilla K8s clusters
Challenges
Create / Destroy k8s cluster
Easily, quickly
How: Creating a cluster
- Build an Koos image
+ k8s Optimized OS
+ Based on Chromium OS - Create 1 Master VM
+ Configure using Metadata server - Create Nodes VM
+ In an MIG (Auto-scaling groups)
+ Create a NodePool (that contains the MIG) - Networking Magic
+ Overlay network for the Pods
+ Routes to reach the master
=> 3 min
How: Upgrading
Monitoring:
* Watch the master VM
* Watch the Kube-system
* Watch the nodes VM
* Watch the MIG
Notes:
* ECTD is store on persistent disk
* Master has STATIC IP
Process:
* Master:
+ Shutdown the master
+ Restart the master with new Image base
* Nodes:
+ Update MIG template
+ Pick a node and drain it
+ MIG recreate a new Node according to the MIG template
OR: Use the NodePool feature
+ New Pool with same / new MIG template
+ Blue/green style
How: Repairs
?
Running you Apps
Leverage GKE Metadata as labels
Auto-scaling
Pods: OK
Nodes: SIG
* Cluster autoscaling is a signal that k8s will emit
* Cloud provider will have to support it
Global network: AKA federation
Federated Ingress !
Soon