Is there anybody out there?
summary
Covers concrete failure detectors and membership implementations used with Akka Cluster: phi accrual and a simpler adaptive variant, SWIM with indirect probes and the memberlist Go implementation (optimizations drop false positives to under 2%), and Rapid which stabilizes clusters >1,000 nodes even under packet loss. Explains gossip styles and piggybacking, shows vector clocks/version vectors for merging, describes CRDTs for strong eventual consistency, and details cluster features like leader-by-convention, cluster-singleton and sharding with in-memory entities for low latency.
AI-written summary · about
chapters (14)
- 0:00Start
- 1:46Motivational quote
- 3:34Key issues for building clusters
- 4:33Implementing group membership
- 5:35Failure detector key
- 7:10Trade-offs
- 8:28Phi Adaptive Accrual
- 11:10SWIM Failure Detector
- 12:58Lifeguard Failure Detector
- 14:17Rapid
- 18:45Gossip styles
- 20:24What do you even gossip about
- 25:14Reaching consensus: time
- 40:31Akka Cluster: sad path (aka reality)
related talks
29mIS THERE ANYBODY OUT THERE?
41mHands on Introduction to Distributed Systems Concepts with Akka Clustering
30mCONSCIOUS CONSISTENCY WITH AKKA CLUSTER, CRDTS AND DISTRIBUTED DATA
31mAKKA CLUSTER MANAGEMENT AND SPLIT BRAIN RESOLUTION
25mGo distributed (and scale out) with Actors and Akka Clustering
01h 59mAkkaCluster and AkkaTyped DeepDive Into the ActorModelWith a BitofEyeCandy.
37mAkka Present and Future
52mDistributed Computing Dealing with Time and Failure in the Wild