Taming the Modern Public and Private Clouds with Nomad
summary
An open‑source cluster scheduler called Nomad claims to run heterogeneous workloads across regions and data centers with a single binary and a simple HCL job spec, using drivers for Docker, runc, VMs and Windows; it does bin‑packing, job priorities, node drain for maintenance, system jobs and Consul‑backed service discovery. Architecture is event‑driven and concurrent (inspired by Omega) and the project reports 1M containers in under 5 minutes and allocation dispatch in ~7–10 ms; current gaps include automated preemption and guaranteed database workflows.
AI-written summary · about
chapters (19)
- 0:00Introduction to Nomad
- 1:48Modern cloud infrastructure challenges
- 8:09Microservices and distributed systems
- 10:58Why we need cluster schedulers
- 13:03Nomad architecture overview
- 21:10Job specifications and drivers
- 25:27Nomad server and client modes
- 29:58Gossip and Raft in Nomad
- 31:00Event-driven scheduling model
- 34:30Q&A on Nomad server
- 35:55Client OS and driver support
- 39:47Favorite Nomad features for SREs
- 44:01Service discovery and logging
- 46:31Production readiness of Nomad
- 49:20Live demo setup
- 56:09Running Nomad server and client
- 57:35Deploying a job with Nomad
- 59:42Debugging failed allocations
- 1:00:50Closing and Q&A
related talks
27mNomad: Past, Present, and Future
24mWhat it Takes to Build & Operate a Global Serverless Provider
27mIntro: Virtual Kubelet
28mDeploying a Multi-Cloud MMORPG with Terraform, Nomad and Consul
26mTech-Nomadic, Run Your Software Anywhere
47mCloud Native and Sustainable, Reproducible Scientific Computing
27mFrom Supercomputing to Serving: A Case Study Delivering Cloud Native Foundation Models
43mCloud Native - effective computing at scale