Writing Stateful Frameworks
summary
Mesos now provides persistent volumes and dynamic reservations but writing failover-safe stateful schedulers remains hard — an initial Cassandra framework grew to 9000 lines of C++ handling reservations, volume recreation and placement. DC/OS Commons SDK generates scheduler logic from a YAML or Java spec, creating plans and phases (serial journal/name-node ordering, parallel data-node rollouts), built-in recovery strategies, and a Hello‑World service that deploys with persistent volumes and works in demos; Uber and others run SDK-based services in production.
AI-written summary · about
chapters (6)
- 0:00Start
- 5:03Framework Scheduling
- 26:35Failure Recovery
- 31:24Deployment Plan
- 34:12Placement Constraints
- 43:20What's Next
related talks
38mState is Hard: An SDK for Building Stateful Applications
59mBuilding Distributed Systems with Apache Mesos
34mPatterns for building Highly Available Cluster Schedulers
33mDistributed Scheduler Hell
27mKeynote Panel: Future of Cluster Management
21mDDD and FSM: tackling complexity with state machines
27mEmpowering Resilient Scalability: The Power of Choreography in Distributed Systems
25mElastic Scaling of Stateful Network Functions