Hi, Your best bet would be to spread the ZK servers among odd number of groupings. e.g. spread them between Datacenter A,B and C so you can take down an entire DC for maintenance. If you don't have another DC available, maybe the data center is split with clouds so you can bring down one cloud in a DC and leave the other one running. You get the idea. Essentially, if you have N/2 and N/2 nodes in 2 DCs its impossible to have a functional ZK when you take down one DC.
We are running 3,3,3, in 3 datacenters. 9 is the smallest number where we could get a redundant ZK ensemble. 2,2,2 config can only handle failure of 2 nodes. -Ayon On Fri, Jan 9, 2015 at 12:50 AM, Kambiz Niktabar <[email protected] > wrote: > Hi, > We are using ZooKepper ensembles to manage configuration files of our Solr > environment. Basically our servers are distributed to two different > physical locations (e.g. Site A and Site B) which every location has its > own maintenance window for security patching, etc. That means during the > maintenance time, it might happens that all servers in site A will be > unavailable.Since the majority of ZooKeeper instances needs to be up and > running to have ZooKeeper available, this forces us to perform maintenance > on the site with minority instances manually.I wonder if there is any smart > way to setup Zookeeper to have fully redundant service for these two > location without involving any manual job?please advise > RegardsKambiz Niktabar
