Thanks a lot Harsh. This is exactly what I did. I appreciate it for having the time to point the correct path as well.
Cheers, Francisco On Sat, 15 Sep 2018 at 19:21, Harsh J <[email protected]> wrote: > As mentioned by Wellington, the path you're going down will not > guarantee any compatibility in future. > > The format/encoding/content/location/etc. of the messages stored in ZK > by the Failover Controller is not for public access and can change > without formal deprecation/etc. > > A cleaner way without the use of commands could be to simply query the > /jmx JSON and parse out the state, something like the below done with > 'jq': > > ~> curl http://some-nn-host:webport/jmx | jq '.beans[] | > select(.name=="Hadoop:service=NameNode,name=FSNamesystem") | > .["tag.HAState"]' > > That said, the protocol-buffers structure used to encode the bytes in > ZK is defined in the following file: > > https://github.com/apache/hadoop/blob/trunk/hadoop-hdfs-project/hadoop-hdfs/src/main/proto/HAZKInfo.proto > > You can use protoc with --go_out to compile that message into a struct > definition, and then use it in a ZK program like this: > https://gist.github.com/QwertyManiac/d3ff72a29e6defd59c353b8b6ca70418 > > On Fri, Sep 14, 2018 at 1:30 PM Francisco de Freitas > <[email protected]> wrote: > > > > Hi Harsh, thanks for your message. > > > > The thing is that I need to find that out via API programatically so > that's why I gave the zkCli command as an example. I'm using this Go Lib ( > github.com/samuel/go-zookeeper/zk) and I get the same result. > > > > On Fri, 14 Sep 2018 at 02:22 Harsh J <[email protected]> wrote: > >> > >> The value you are looking at directly in ZooKeeper is in a > >> serialized/encoded form. Those are not separator characters but more > >> likely an encoded integer binary value that your terminal is > >> interpreting as a printable character. > >> > >> The standard way to find the active NameNode is to use the 'hdfs > >> haadmin -getAllServiceState' command: > >> > >> [hdfs@host ~]# hdfs haadmin -getAllServiceState > >> host1.com:8022 standby > >> host2.com:8022 active > >> > >> You can then extract out just the active NameNode hostname: > >> > >> [hdfs@host ~]# hdfs haadmin -getAllServiceState | grep active | awk > >> -F: '{ print $1; }' > >> host1.com > >> On Thu, Sep 13, 2018 at 9:39 PM Francisco de Freitas > >> <[email protected]> wrote: > >> > > >> > When querying different HDFS clusters I get different separators > (don't really know if they're actually separators). > >> > > >> > From zkCli.sh on different clusters, running the following I get: > >> > > >> > cmd: get /hadoop-ha/clusterX/ActiveBreadCrumb > >> > > >> > Cluster1 (comma): > >> > cluster1active-nn1,active-nn1.example.com �>(�> > >> > > >> > Cluster2 (single double quote): > >> > cluster2active-nn2"active-nn2.example.com �>(�> > >> > > >> > Cluster3 (dollar sign): > >> > cluster3active-nn3$active-nn3.example.com �>(�> > >> > > >> > How can I effectively write a generic code deployed on different HDFS > clusters to effectively find out which is the active NN from querying ZK? > >> > > >> > Or am I doing something wrong? Is the behavior above expected? > >> > >> > >> > >> -- > >> Harsh J > > > > -- > Harsh J >
