IS THERE ANYBODY OUT THERE?
summary
Cluster membership is implemented using concrete tools: Phi accrual (and a JVM-tuned adaptive variant) for failure detection, SWIM with SWIM + Lifeguard / memberlist to cut false positives, random gossip with piggybacking and seen-tracking for dissemination, and vector clocks plus CRDTs for conflict resolution. Akka Cluster measures RTTs by ping‑pong to five peers, adapts suspicion thresholds for cloud GC/noisy networks, elects leaders by deterministic ordering, and relies on leader-driven membership decisions and separate split‑brain resolvers to handle partitions.
AI-written summary · about
chapters (16)
- 0:00Disclaimer
- 2:46Motivational quote
- 4:14Key issues for building clusters
- 5:07Implementing group membership
- 5:32Failure detector key properties
- 6:52Trade-offs
- 7:25Phil Adaptive Accrual
- 8:46New Adaptive accrual
- 9:16SWIM Failure Detector
- 11:15Lifeguard Failure Detector
- 12:38Dissemination strategies
- 14:10Gossip styles
- 16:59Reaching consensus: time
- 23:36Reaching consensus: conventions
- 27:38Akka Cluster: sad path (aka reality)
- 28:39Thank you!
related talks
45mIs there anybody out there?
31mAKKA CLUSTER MANAGEMENT AND SPLIT BRAIN RESOLUTION
41mHands on Introduction to Distributed Systems Concepts with Akka Clustering
25mGo distributed (and scale out) with Actors and Akka Clustering
30mCONSCIOUS CONSISTENCY WITH AKKA CLUSTER, CRDTS AND DISTRIBUTED DATA
44mLITHIUM: AN OSS SPLIT-BRAIN RESOLVER FOR AKKA-CLUSTER
01h 59mAkkaCluster and AkkaTyped DeepDive Into the ActorModelWith a BitofEyeCandy.
44mCluster Consensus: When Aeron Met Raft