Hi, I think this is accomplished with metadata requests. A client sends this request (along with a list of topics it's interested in) to a broker in the cluster and gets back a list of leader/replica partitions. This is cached and refreshed, and when a client receives the "Not a Leader" error (means a request was sent to the wrong broker or a broker that went offline) they'll request the metadata again to find out who the new leaders are.
My source for this is Kafka The Definitive Guide, Chapter 5, but someone correct me if this is wrong. Harper On Tue, Sep 24, 2019 at 11:15 PM Isuru Boyagane < isuruboyagane...@cse.mrt.ac.lk> wrote: > Hi, > > Can anyone clarify the procedure of a client getting to know about a Kafka > leader failover and identify the new leader? > > Thanks. >
