How Uber Built Odin to Handle 3.8 Million Containers

Uber developed Odin, an automated, technology-agnostic platform for managing 3.8 million containers and 300,000 stateful workloads across 100,000+ hosts. Odin replaced manual database management with declarative automation, self-healing remediation loops, and dynamic resource scheduling, enabling zettabyte-scale storage management for services like ride-hailing and payment processing. Key innovations include make-before-break migrations, colocated databases, and a global coordination system for fault tolerance.

---

Core Technical Concepts/Technologies

• Declarative state management (goal-driven automation)
• Self-healing remediation loops (Kubernetes-inspired)
• Grail: Real-time global infrastructure monitoring
• Cadence workflows (orchestration)
• Containerized stateful workloads (100 databases/host)
• Make-before-break migration strategy
• Host-level agents (Odin-Agent + tech-specific workers)
• Support for 23+ storage systems (MySQL, Cassandra, Kafka, HDFS)

---

Main Points

• Scale:
  • 100,000+ hosts, 3.8M containers, 300K workloads
  • Zettabyte-scale storage (multiple exbibytes)
• Automation:
  • Declar