Fault Domains in Mesos
summary
Mesos 1.4 adds a native fault domain primitive (stored in domain_info protobuf) with a simple region/zone hierarchy and offers that include an agent’s domain so frameworks can place tasks by domain. Frameworks opt in via a region-aware capability to receive remote-region offers; non‑opted frameworks don’t get them. Use cases: hybrid cloud bursting, rack-aware placement for stateful replicas (example: uniformly spread nine shards across three zones), and upgrade rules require masters first; feature ships experimental in 1.4.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:25About me
- 2:15Use case #1: Fault tolerant scheduling
- 3:21Bad scheduling
- 4:33Use case #2: Hybrid Cloud
- 6:13Hybrid Cloud Scheduling considerations
- 8:23Existing solutions
- 10:43Goals
- 11:59Solution Overview
- 13:45Fault Domain Hierarchy
- 17:42Terminology
- 19:37Implementation details
- 20:04Master changes
- 21:09Agent changes
- 22:29Framework changes
- 23:58Examples with Marathon
- 25:42Upgrades
- 29:31State of the feature
- 30:46Acknowledgements
related talks
35mFault Tolerant Frameworks - Making Use of CNI Without Docker
25mExplore fault isolation boundaries on AWS
59mBuilding Distributed Systems with Apache Mesos
40mDIY Fault‑Tolerant Microcluster
21mDDD and FSM: tackling complexity with state machines
19mCAFault: Enhance Fault Injection Technique in Practical Distributed Systems via...
22mMonarch: A Fuzzing Framework for Distributed File Systems
43mCreating Fault Tolerant Services on Mesos