Avoiding single point of failure in a multi-services architecture
summary
Plivo runs 100+ servers serving 100 million minutes of voice and avoids single points of failure by separating stateless (Flask, Django, Celery) and stateful (FreeSWITCH media servers on hardware) tiers, using OpenSIPS for SIP, Postgres + Redis for storage, Sensu for monitoring and SaltStack + Jenkins for automated orchestration. Automated failover promotes Postgres slaves and Salt pushes new DB client configs across the cluster (faster than DNS), auto-scaling replaces stateless nodes with no human intervention, and a custom P infra tool exposes safe role-based commands.
AI-written summary · about
chapters (10)
- 0:00Start
- 0:49WHAT WE DO
- 1:05OUR SCALE
- 1:27OUR UPTIME
- 2:07OUR STACK
- 7:08OUR ARCHITECTURE
- 10:28ARCHITECTURE IN PRACTICE
- 12:38HOW WE GOT STARTED
- 14:22HOW IT LOOKS TODAY
- 22:02THANK YOU
related talks
31mEscaping operations hell
42mBuilding a hosting platform with Python
38mBuilding a high performing Passenger Information System
42mPlanning To Fail
18mScaling, what you need to know
43mFast and Furious: Startup scaling with Kubernetes
14mThe Multi‑Workload Challenge: Predictable Failover
46mPlatforms at Twilio: Unlocking Developer Effectiveness