AnsweredAssumed Answered

Understanding CLDB, Primary Region Selection, ZooKeeper etc - from a process perspective

Question asked by rthimmana on Jun 5, 2013
Latest reply on Jun 6, 2013 by Ted Dunning

We install multiple CLDBs and ZooKeepers (in odd numbers) to have hot stand-by. Is there some documentation that explains the best practices

We put a 4 node cluster, pre-split our tables, had 4 CLDBs and 1 ZooKeeper running on our cluster. Started loading to get some stats on the number of operations per sec on M7 tables. The data seemed to be well distributed between the nodes of the cluster - wrt the primary-region / node ratio

During subsequent runs, we got some alarms for the file server on a couple of nodes. Changed the memory settings for the file server and restarted the wardens, while the loads were still continuing. Now, the majority of the primary-regions are on node1, a small % on node2 and 0% on nodes 3 and 4

So, all our map reduce jobs are running on 1 core on node1. Are there any documents that describe the workings of the CLDB and the ZooKeeper and the factors on which the Primary-Region is selected and if this can be changed / balanced to ensure maximum resource usage on the Cluster?