

20160615 2 Containers swap

20160615_2_Containers-swap

• Principles
• Bootstapping etcd

Principles

Build OS images
- Testable
- Repeatable
- Faster to instantiate
- More reliable (Time to install packages, remote dependencies …)

Tips:
* Use a "minimale" OS image

Cluster all the things
- No single point of failure

Bootstapping Etcd
- Discovery service ?
+ DONT depend on the Public service
+ Host your own private one
=> BUT still something to maintain … while bootstrapping ?
- DNS
+ Too statics

=> Etcd bootstraping service:

Bootstapping etcd

Shell script running on each CoreOS instance
- Relying on AWS Auto-scaling feature

* Determin if a cluster exists
  - query local ETCD running
* If not, start one
  - query AWS autoscaling group
  - get list of existings nodes
  - Wait till there are enough members to create a cluster
* Otherwhise, join
  - query AWS autoscaling group
  - get list of existings nodes

=> “External” Metadata store is the Key !!!