about:infrastructure

Transcript

1. about:infrastructure https://github.com/yyuu
2. Outline 1. What’s the infra engineer? 2. What’s the fault torelant system? 3. Appendix: the current & future of the fault torelant systems
3. 1. What’s the infra engineer? A person who build and maintain the “infrastructure”.
4. 1-1. What’s the infrastructure? Keep everything up and running: * Design * Implementation * Maintainance ⇒ It’s all based on the design. High availability =~ Fault torelance
5. 2. What’s the fault torelant system? Make faults ignorable. 1. Redundancy 2. Homeostasis
6. 2-1. Redundancy Allocate enough capacity including spairs. * N+M Redundancy * Have at least N with M spairs for faults. * Active/Passive (Active/Backup) * Single master with hot standby
7. 2-1. Redundancy (examples) * N+M Redundancy: * LB + app servers * RAID5/RAID6/RAID-Z * Active/Passive: * MySQL replication * HSRP/VRRP/UCARP ⇒ There’re the best practices
8. 2-2. Homeostasis Keep a process up and running. * Health-check * Recovery
9. 2-2. Homeostasis (examples) * Health-check & Recovery: * mon, monit * cron * chef * puppet ⇒ continuous process to detect faults ⇒ need automation
10. Conclusion 1. Learn the best practices * Try AWS/GCP & Cloud Design Patterns * Redundancy is by design; kill all SPOFs 2. Automate everything =~ programming * Cloud services must have its API * Design your API and automate the process
11. Appendix: the current & future of the fault torelant system
12. The fault torerant system current: * Redundancy * Homeostasis future:
13. The fault torerant system current: * Redundancy * Homeostasis future: * Redundancy * Homeostasis * Apoptosis (NEW!)
14. Apoptosis: programmed suicide * The death of long living processes have great impact to the overall system. ⇒ If so, kill them all before it’s getting old... * Netflix/SimianArmy (Chaos Monkey) * Terminate EC2 instances in random manner
15. To use “apoptosis” of the system... * Make the lifecycle of processes shorter * Aggressive system-wide “homeostasis” * GoogleCloudPlatform/kubernetes * Decouple the app and the system * Use containers =~ Docker
16. EOF