Nomad: Scheduling Applications at Scale
summary
Nomad is a HashiCorp scheduler that runs as a single binary install, uses Consul/Serf for gossip and consensus across regions and data centers, and accepts human-friendly JSON-like job files that describe resources and constraints. It schedules Docker, rkt, KVM and standalone binaries, enforces priorities and bin‑packing, and reschedules on node failures. Architecture is optimistically concurrent with client-side fingerprinting. The claim is proven by the million container challenge: over 1,000,000 containers on 5,000 hosts scheduled in ~267 seconds, with automatic recovery.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:27Standard Schedulers
- 5:13CPU Schedulers
- 11:55Other Schedulers
- 15:13Design Architecture
- 23:16Evaluations
- 25:23Constraints
- 26:56Flexibility
- 29:10Negative
- 31:15Demo
- 35:18Questions
related talks
01h 03mScaling Web Operations with Distributed Cluster Schedulers
33mDistributed Scheduler Hell
26mNomad: Past, Present, and Future
30mA beginner's journey: Scaling self-hosted platform infrastructure with Nomad
34mPatterns for building Highly Available Cluster Schedulers
37mDistributed Flexibility: Nomad and Vault in a Post-Kubernetes World
30mHashiCorp's Game of Life
25mAutoscaling Your Nomad Jobs Using SRE Insights