Fault Tolerant Frameworks - Making Use of CNI Without Docker
summary
A fault‑tolerant Mesos framework and a Go SDK that use the v1 streaming API and the UCR (universal containerizer) with CNI networking replaces Docker and marathon for smaller clusters. The implementation uses protobufs, leader election via TTL keys backed by etcd (raft), and a custom executor; it launches 500 tasks in a live demo and was able to push 50,000 tasks in ~40 seconds, consuming as little as ~15 MB of memory for the scheduler. Mesos <1.2 UCR bugs are skipped and multiple NICs per container are supported.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:28Choosing an API
- 1:35Using Go
- 3:11Go SDK
- 5:10Docker
- 5:52Why use the UCR
- 7:17UCR Issues
- 9:38Live Demo
- 10:05Launching Tasks
- 14:42Open Source
- 18:23CNI
- 20:07TCP Connection
- 21:27NCD
- 23:34Logs
- 26:12Why we made our own
- 28:56Questions
related talks
33mContainer Networking for Micro-Services: An Apache Mesos Networking Deep Dive
46mContainer Network Interface Onwards and Upwards
46mWriting Stateful Frameworks
31mDistributed system fault injection testing with Docker
58mDistributed Scheduling with Apache Mesos in the Cloud
43mFault Domains in Mesos
39mProcess Migration in the Orchestration World
52mNightmares of a Container Orchestration System