The art of service discovery at scale
summary
Describes Netflix’s Eureka service discovery: a heartbeat + registration protocol that favours availability over consistency, replicates ephemeral instance state across nodes, and tolerates stale entries to survive partitions. Shows concrete failure modes: missed heartbeats (90s eviction window), split-brain partitioning, and how self‑preservation prevents mass eviction. Presents a coordination‑free design that replicates ordered event streams between discovery nodes, gives bounds on convergence time, and reports that the approach works in production to keep autoscaling and client lookups functional during datacenter faults.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:52What is service discovery
- 3:30How a service discovery used
- 10:49What if service discovery is unavailable
- 12:43Cap Theorem
- 14:21Availability
- 15:27Durability
- 18:27Stateful protocol
- 19:08Stateful interaction model
- 21:41Data conflicts
- 28:16Cost of divergence
- 30:04Ordered stream
- 30:49Replication
- 31:39Network partitions
- 32:37Selfpreservation
- 33:55Impact of failure
related talks
46mExperiences from using discovery services in a microservice landscape
01h 09mKey Problems of Microservice Architecture
40mCloud Native Java
27mMicroservices: Awareness and Discovery Pattern on K8s
02mHablamos con Mark Heckler
28mHow Netflix Makes gRPC Easy to Serve, Consume, and Operate - Benjamin Fedorka, Netflix
35mCaching at Netflix: The Hidden Microservice
46mReliability Nirvana