Hey Danny Currently the bootstrap servers are only used when the client initialises (there’s a bit of discussion around the issue in the jira below if you’re interested). To implement failover you’d need to catch a timeout exception in your client code, consulting your service discovery mechanism and reinitialise the client.
KAFKA-3068 <https://issues.apache.org/jira/browse/KAFKA-3068> B > On 31 May 2016, at 22:09, Danny Bahir <dannyba...@gmail.com> wrote: > > Hello, > > Working on a multi data center Kafka installation in which all clusters have > the same topics, the producers will be able to connect to any of the > clusters. Would like the ability to dynamically control the set of clusters a > producer will be able to connect to, that will allow to gracefully take a > cluster offline for maintenance. > Current design is to have one zk cluster that is across all data centers and > will have info regarding what in which cluster a service is available. > > In the case of Kafka it will house the info needed to populate > bootstrap.servers, a wrapper will be placed around the Kafka producer and > will watch this ZK value. When the value will change the producer instance > with the old value will be shut down and a new producer with the new > bootstrap.servers info will replace it. > > Is there a best practice for achieving this? > > Is there a way to dynamically update bootstrap.servers? > > Does the producer always go to the same machine from bootstrap.servers when > it refreshes the MetaData after metadata.max.age.ms has expired? > > Thanks!
