Hi Gordon, To pile on what Cos said, YARN's ResourceManager also stores it's state in ZK. See the YARN JIRA <https://issues.apache.org/jira/browse/MAPREDUCE-279>and design doc<https://issues.apache.org/jira/secure/attachment/12486023/MapReduce_NextGen_Architecture.pdf>for more details. And, adding a dependency like this also allows us to harmonize the version of zookeeper (which is increasingly becoming a fundamental building block) being used by various components in the ecosystem.
Mark On Mon, Mar 17, 2014 at 9:34 AM, Konstantin Boudnik <[email protected]> wrote: > Hadoop2 HDFS HA uses ZK for all sorts of things, hence the dependency. > > You are right that none of the Hadoop core components require ZK for normal > operations. But once HA is needed HDFS will be needing that. > > Hope it helps, > Cos > > On Mon, Mar 17, 2014 at 04:26PM, Gordon Wang wrote: > > Hi guys, > > > > Could you please help me to understand why hadoop rpm depends on > zookeeper > > in bigtop rpm specs ? > > > > IMO, the hdfs-fc should depend on zookeeper. But for NameNode, DataNode, > > ResourceManger, NodeManger, I think they should work without zookeeper. > > > > > > Thanks. > > > > -- > > Regards > > Gordon Wang >
