James Il Lun 7 Set 2020, 18:07 James Gover <[email protected]> ha scritto:
> Hi, > > We are upgrading from zookeeper 3.4.8 to 3.5.8 via 3.4.14. > There are several mentions of going from 3.4.6 to 3.5.8, is going from > 3.4.14 to 3.5.8 ok. > > In terms of Leader-tasks running during the upgrade. The intent is to > migrate a cluster of 3 zk nodes by: > * scaling up to 5 nodes (adding two new nodes) > * one non-leader node at a time; destroying the instance and replacing it > with a new 3.5.8 node > * then stoping zk on the leader so another node becomes the leader, > terminating the instance then adding a 3.5.8 node. > * removing two (non-leader) node to scale back to 3 nodes > > With this process; will leader tasks running on the 3.4.14 leader (using > curator leaderLatch) fail when this leader is terminated but succeed if run > after a short delay will run on a new 3.5.8 leader node. > Zookeeper itself does not run tasks, it is only the coordination system that tells which one is the current leader among the peers. The upgrade of Zookeeper should be seamless for the application as far as you keep the Zookeeper quorum up and running, usually by keeping alive half+1 number of Zookeeper nodes. I hope that helps Enrico > Thanks, > > James > > > > -- > Sent from: http://zookeeper-user.578899.n2.nabble.com/ >
