Automatic Datacenter and Service Deployments...
summary
Alibaba builds Apsara Stack as an image-driven, immutable bare-metal infrastructure with a central image builder, per-datacenter image caches and local downloader agents; operating systems come from a provisioning service that reads YAML configs and an in-OS agent (Osman). Capacity planning splits into business and delivery phases and artifacts model product/service/application plus placements. A stateful controller (about 5–6 states) moves deployments and cuts bootstrap time from months to typically 24–72 hours.
AI-written summary · about
chapters (19)
- 0:00Automatic datacenter and service deployments based on capacity planning artifacts
- 1:02How to deploy a datacenter for cloud service?
- 1:56Difficulties to deploy a datacenter
- 4:32Our Solution
- 5:07Traditional Deployment Method
- 8:43Better solution: let's lock it down
- 11:32Immutable Infrastructure: Image based Deployment
- 12:48Image building process
- 15:05Normal immutable infrastructure
- 17:09Immutable intrastructure in a bare-metal environment
- 20:57Our approach to deploy a datacenter with cloud service Business planning phase Delivery planning phase
- 23:00Concepts in the artifacts
- 25:50Product - Services
- 28:50Product Features
- 31:24Service - Server Roles
- 32:01Service - Dependencies
- 34:34Application
- 35:12Data center described in artifacts
- 38:06Summary
related talks
32mNo Touch Installation Concept
14mAMI Creation And IaC Deploys via Semaphore
01h 01mTaming the Modern Data Center
25mAutomatic Provisioning of Bare Metal Edge for AI Inference
22mLeveraging Chef, Packer, and Habitat for Effortless AMI Deployments
44mThe Amazon Software Development Process
42mCatastrophes and Atrophies
38mDon’t Ever Change! Are Immutable Deployments Really Simpler, Faster, and Safer?