Hi Kevin, It is possible to have multiple Namenodes in a HA setup. One Active Namenode and multiple Standby Namenodes. The Active Namenodes serves all the requests. In case of failure of Active Namenode, one of the standby takes up its place and acts as Active to serve the requests there after. At any time there would be only one Active Namenode in a cluster.
You can give a check here to know about HA: https://hadoop.apache.org/docs/r3.2.0/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.html If you intend to have multiple namenodes that can serve requests together in similar manor, that isn’t possible, But in Trunk code, we have Observer Namenodes that can serve the Read Requests but not the write Requests. The Write Request can be handled only by Active Namenode. You can check stuff regarding Observer Namenode here: https://issues.apache.org/jira/browse/HDFS-12943 -Ayush > On 08-Jun-2019, at 10:31 PM, Sunil Govindan <sun...@apache.org> wrote: > > Hi Kevin > > Yes, its possible. Please refer to > https://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/ResourceManagerHA.html > This explain how to configure RM in HA mode when one can be standby if that > what you are looking for. > > Each NM sends a heartbeat like in every second to RM, and RM will do > allocation based on free resources available in these nodes during heartbeat. > > - Sunil > > >> On Sat, Jun 8, 2019 at 6:30 AM kevin su <pings...@gmail.com> wrote: >> Hi all, >> >> There is a way to start two Namenode, and request for both of them >> >> So can we start two RM as well ? >> >> Another question is every times if we want to allocate resource from RM, it >> will find which node manager has resources. >> >> Is RM send every request to all node manager, and get all NM report, or >> there is a meta-data in RM, so it will directly find one of NM to get >> resources ? >> >> Thanks in advanced >> >> Kevin >> Best regards