Creating Fault Tolerant Services on Mesos
summary
Shows how ArangoDB's Raft-based Agency provides a central, transactional key/value store (JSON HTTP/REST) that implements transactions with preconditions, TTLs, and HTTP callbacks, plus supervisor code run in the leader. Demonstrates concrete patterns: service discovery by registering addresses, one-time initialization using TTL + conditional write, and leader election via TTL renewal. Includes runnable examples and a small repository with ~100-line implementations that run in containers, claimed to be robust in practice.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:30Algorithms Methods Tools
- 4:50Traditional Consensus
- 6:40Advice
- 12:40Key Value Store
- 14:40Run Implementation
- 15:57Philosophy
- 16:57Transsexual Store
- 17:53State Projection
- 19:55Right Transactions
- 22:22preconditions
- 23:59subscribe
- 24:49caveat
- 25:47supervision
- 28:49typical use case
- 31:16discovery
- 33:29onetime initialization
- 34:46failure of initialization
- 39:28leader update
related talks
40mDistributed Data on Distributed Infrastructure
47mDistributed Systems: When Democracy and Computer Science Come Together
38mDisaster Recovery with a Distributed Database
50mDo Distributed Systems they said. It'll be fun
41mThe Raft Consensus Algorithm
43mBuild Your Own Distributed System Using Go
12mFault-Tolerant Replication with Pull-Based Consensus in MongoDB
43mRa: a Raft implementation