Eventually Consistent Service Discovery
Talkat SREcon19 Europe/Middle East/Africa · on Wednesday, 2019-10-02· at 14:30· Track 1: The Liffey B·
summary
Argues that service discovery usually tolerates eventual consistency and that Monzo implements an Envoy sidecar model with a config provider backed by gossip/SWIM rather than forcing Raft/etcd for every change. Shows problems with strong-consensus: leader bottlenecks, disk commit latency, and scalability limits, and with gossip: quadratic message growth and propagation lag. Cites SWIM to mitigate costs, reports 1,300 microservices and thousands of instances, and claims Envoy converges in seconds.
AI-written summary · about
chapters (8)
- 0:00The legacy of manual locking
- 1:43Introduction to service discovery
- 2:37Strongly consistent systems
- 8:52Scaling and performance limits
- 10:23Sequential consistency
- 12:19Eventually consistent systems
- 13:45Optimizing gossip and discovery
- 23:01Conclusion and Q&A
related talks
27mData Consistency in Microservices Architecture
25mThe shell game called eventual consistency
44mCluster Consensus: When Aeron Met Raft
28mFrom Custom to Community: Embracing Envoy for Service Mesh Proxies at ING
27mPanel: Tales from the Edge: Is the Edge More Important Than the Service Mesh?
38mEventual Consistency: A Question of Time and Expectations
38mTowards consensus on distributed consensus #bbuzz
14mDistributed Consensus in 15 Minutes!