Large-Scaled Deploy Over 100 Servers in 3 Minutes
summary
Company runs Ruby on Rails minne with ~200 servers across 2 data centers and designs an image‑based pipeline to turn mutable infrastructure into immutable infrastructure. A pull deploy strategy using capistrano consul and stretcher plus image creation, a build server for Rails assets and bundle packages, and event-driven registration lets clusters launch and fetch releases; blue‑green switching minimizes downtime and the system completes a 100+ server rollout in about three minutes.
AI-written summary · about
chapters (47)
- 0:00Start
- 0:31self.introduce
- 2:52Call for Speakers
- 4:51Concerns of bootstrap instructions
- 5:25Web operation is manual instructions
- 6:08Background of "No SSH"
- 6:57Provision with puppet
- 7:57To use puppetmasterd . We choice master/agent model . It's large scaled architecture because we didn't need to deploy
- 8:42What's cloud-init
- 9:30Basic usage of cloud-init
- 11:46Upgrading Rails 4
- 12:25What's new for capistrano3
- 12:49Do not use hostname/ip dependency
- 13:37Bundled package of Rails application
- 14:30Distributed rails package
- 15:01Integration of image creation We extracted rails bundle when instance creates self image with
- 15:32How to test instance behavior
- 16:52Scale out with cli command
- 17:24How to automate instructions
- 18:17Concerns of bootstrap time
- 19:03Check point of Image creation
- 19:212 phase strategy
- 20:41Use-case of Packer
- 20:57inside image creation with Packer
- 21:40minimal image
- 22:47web application image
- 25:26What's blocker for scale-out
- 26:58consul + consul-alert
- 28:34Configuration of mackrel
- 29:21access_log aggregator with td-agent
- 30:31Remove to batch scheduler We need to use batch role for scheduled rake task. We have to create some payments transaction, send promotion mail, indexing search items and more
- 31:13scheduler architecture
- 32:18Container based CI with Rails
- 33:31Serverspec
- 33:53Refactoring puppet manifets
- 34:19Switch Scientific Linux 6 to Centos 7
- 34:49All of processes under the systemd
- 36:45capistrano-strecher
- 37:25Architecture of pull strategy deployments
- 38:26Why we choose OpenStack?
- 39:19yaocloud and tool integration We made Ruby client for OpenStack named Yao.
- 40:01Multi DC deployments in 3 minutes
- 40:53Instructions of Blue-Green deployment
- 41:33Dynamic upstream with load balancer
- 42:44Slack integration of consul-template
- 43:03Example code of thor
- 43:58Summary • We can handle TV CM and TV Show used by scale-out servers.
