Power Loss Siren: Making Facebook Resilient to Power Outages
summary
Power Loss Siren is a rack level alerting system that uses existing in rack batteries to detect outages and notify services via a relay daemon on the top of rack switch and handlers on servers. Batteries provide ~90 seconds of backup and the system uses a 45 second threshold to declare impending loss, triggering a live promotion that cuts MySQL primary failover from “tens of seconds” of visible write outage to effectively transparent. It eliminates the need for dual powered rows and reduces buffer capacity requirements.
AI-written summary · about
chapters (10)
- 0:00Power Loss Siren project
- 1:50Data center power outages
- 3:162017 North Carolina outage case study
- 4:20Power outages impact on Facebook
- 7:51Detecting impending outages
- 10:41Mitigating outages for databases
- 12:49Benefits for stateless and storage
- 13:52Simplifying infrastructure with PLS
- 16:42Key learnings and PLS future
- 18:33Q&A
