I guess checking the number of znodes via JMX could work no ?
On Aug 3, 2011 6:14 PM, <[email protected]> wrote:
>
> Hello,
>
> We have the project built on zookeeper-3.3.3 which starts the cluster,
> just simple wrapper on zk lib.
>
> Every version of the project has its own path to zookeeper's data/tx
> folders. But all configs (number of nodes, client ports, etc) - the
> same.
>
> We release a new version and want to deploy it without stopping the
> cluster.
> Let's say we change from version A to B
>
> We are executing:
>
> 1. stop node 1 of cluster (v. A)
> 2. start node 1 of cluster (v. B)
> 3. stop node 2 of cluster (v. A)
> 4. start node 2 of cluster (v. B)
> And so on.
>
> The data is automatically transferred to the data/tx directories of the
> newly started nodes from the still running nodes of project version A.
>
> The question is: is it possinble to check that the transfer of data
> completed? Is it correct that it starts automatically?
>
> Action 1, 2, 3, 4, ... can be automated but if they are executed too
> fast I suppose it's possible to get the incompleted data on the nodes of
> project's B.
>
> Thank you
> Semyon
>
