[Verse 1] When clients need to know what's going on They still talk to brokers, not the control zone No direct controller contact, that's the rule Metadata requests through brokers, that's the tool The architecture keeps the old client way But underneath, KRaft's here to stay [Chorus] Four implications you should know How the client-side will flow Metadata through brokers still Faster updates, that's the thrill Cluster ID and describe calls KRaft improvements help us all [Verse 2] Propagation latency gets a boost today ZooKeeper's slow path fades away Controller to broker, then to client fast No more waiting for consensus to last The metadata flows like water downstream KRaft makes it smooth, fulfilling the dream [Chorus] Four implications you should know How the client-side will flow Metadata through brokers still Faster updates, that's the thrill Cluster ID and describe calls KRaft improvements help us all [Bridge] Take your kafka-metadata shell Snapshot flag will serve you well Dump the log and trace the flow Topic creation, watch it grow Decode the records, type by type See how KRaft handles the hype [Verse 3] DescribeCluster API reveals the truth Cluster ID shows the living proof Lab three will teach you how to see The metadata log's complexity From creation to the final state KRaft's efficiency you'll appreciate [Chorus] Four implications you should know How the client-side will flow Metadata through brokers still Faster updates, that's the thrill Cluster ID and describe calls KRaft improvements help us all [Outro] Client perspective stays the same But KRaft's playing a faster game Through the brokers, information flows How much faster, now you know
← 2 Apache Kafka Fundamentals | 1 Migration Strategy Overview →