Also, there is a description of what is in META and ROOT in here... http://hbase.apache.org/book.html#arch.catalog
... and it also describes the startup sequencing. On 2/10/12 10:46 AM, "Harsh J" <[email protected]> wrote: >The client does communicate with the master to perform .META. changing >interactions (create/delete tables, etc.). And for the rest, like >locating regions off regionservers and reading off them, the master >isn't touched (afaik). > >The master's work is also more about running/providing cluster >management services for the many region servers than serving clients >all the while. > >See the interfaces HMasterRegionInterface and MasterServices on trunk >to get a view of what all functions the master carries out, on a high >level, and why its role is important though clients may not need it >beyond administrative purposes. > >2012/2/10 yonghu <[email protected]>: >> Thanks for your reply. If the -ROOT- and .META. tables are managed by >> two RegionServer separately. What is the functionality of Master node? >> It only assigns the Region node in the cluster? So, the client only >> needs to contact with these two special RegionServer which contains >> the -ROOT- and .META. tables. If the interaction model is like what I >> said, the client will not contact with master node! If I am wrong, >> let me know. >> >> Thanks >> >> Yong >> >> On Fri, Feb 10, 2012 at 12:47 PM, Harsh J <[email protected]> wrote: >>> The HMaster does not host regions, and the -ROOT- is a region; It is >>> hosted by one of the assigned RegionServers, and its location is >>> registered under ZooKeeper. The -ROOT- region then holds the location >>> of the .META. (Which again, is another region, and is hosted by >>> RegionServers in just the same way). >>> >>> For your ZK question, if you run ZK in HBase's embedded mode, it runs >>> on the HMaster, while if you run it standalone you can choose to run >>> it anywhere. The latter method is what is recommended for production >>> deployments, and its preferable to host it separately on its own >>> boxes. >>> >>> P.s. The JIRA at https://issues.apache.org/jira/browse/HBASE-3171 may >>> simplify this someday :) >>> >>> 2012/2/10 yonghu <[email protected]>: >>>> Thanks! >>>> I know this. I just want to know which nodes store this information >>>> when the client first contact to HBase cluster, HMaster or >>>> RegionServer or a special node in which runs the zookeeper. >>>> >>>> And the other question is whether zookeeper runs on the same nodes as >>>> Hbase in the cluster or it runs in a separate nodes? >>>> >>>> Yong >>>> >>>> On Fri, Feb 10, 2012 at 12:16 PM, Roger <[email protected]> wrote: >>>>> To my knowledge, it is a three level tree-like structure. >>>>> ------------------ >>>>> 该邮件从移动设备发送 >>>>> >>>>> ------------------ Original ------------------ >>>>> From: "yonghu"<[email protected]> >>>>> Date: Fri, Feb 10, 2012 07:12 PM >>>>> To: "user"<[email protected]>; >>>>> Subject: Which server store the root and .meta. information? >>>>> >>>>> >>>>> Hello, >>>>> >>>>> I read some articles which mention before the client connect to the >>>>> master node, he will first connect to the zookeeper node and find the >>>>> location of the root node. So, my question is that the node which >>>>> stores the root information is different from master node or they are >>>>> the same node? >>>>> >>>>> Thanks! >>>>> >>>>> Yong >>> >>> >>> >>> -- >>> Harsh J >>> Customer Ops. Engineer >>> Cloudera | http://tiny.cloudera.com/about > > > >-- >Harsh J >Customer Ops. Engineer >Cloudera | http://tiny.cloudera.com/about >
