Balancer not working in hdfs HA

[#1]

I was working on one of the case and came across an issue where balancer utility was not working in namenode HA environment.

HDP Version - 2.3.2.0

Ambari Version - 2.1.2

Issue - balancer was failing with below error:

[hdfs@slave1 ~]$ hdfs balancer 15/11/20 11:46:10 INFO balancer.Balancer: namenodes = [hdfs://mycluster, hdfs://mycluster:8020] 15/11/20 11:46:10 INFO balancer.Balancer: parameters = Balancer.Parameters [BalancingPolicy.Node, threshold = 10.0, max idle iteration = 5, #excluded nodes = 0, #included nodes = 0, #source nodes = 0, run during upgrade = false] [..] 15/11/20 11:46:11 INFO block.BlockTokenSecretManager: Setting block keys 15/11/20 11:46:11 INFO balancer.KeyManager: Update block keys every 2hrs, 30mins, 0sec java.io.IOException: Another Balancer is running.. Exiting ... Nov 20, 2015 11:46:11 AM Balancing took 1.883 seconds 

Resolution(workaround) -

1. Copy /etc/hadoop/conf/* /home/hdfs/copied_conf/

2. Edit /home/hdfs/copied_conf/hdfs-site.xml and delete below property

<property> 
 <name>dfs.namenode.rpc-address</name> 
 <value>xyz:8020</value>
</property> 

3. Run balancer with below command(as user hdfs):

hdfs --config /home/hdfs/copied_conf balancer 

Permanent Solution -

This will get fixed in Ambari 2.1.3. version (Refer - https://hortonworks.jira.com/browse/BUG-45749)

[#2]

thank you @Kuldeep Kulkarni I have same issue with a prospect. Same happens to hdfs mover.

[#3]

@Guilherme Braccialli - Glad this helped! 🙂

[#4]

I also found this jira: https://issues.apache.org/jira/browse/AMBARI-13946

[#5]

@Kuldeep Kulkarni another workaround:

1- Execute commands below:

/var/lib/ambari-server/resources/scripts/configs.sh -u admin -p AMBARI-PASSWORD delete localhost CLUSTER-NAME hdfs-site "dfs.namenode.rpc-address"
/var/lib/ambari-server/resources/scripts/configs.sh -u admin -p AMBARI-PASSWORD delete localhost CLUSTER-NAME hdfs-site "dfs.namenode.http-address"
/var/lib/ambari-server/resources/scripts/configs.sh -u admin -p AMBARI-PASSWORD delete localhost CLUSTER-NAME hdfs-site "dfs.namenode.https-address"

2- Restart HDFS

[#6]

Thanks @Guilherme Braccialli

[#7]

Great! I believe ambari and hdfs services restart is needed if we delete configs using configs.sh script.

[#8]

@Kuldeep Kulkarni ambari restart is not needed, I tested and updated configs appears immediately. You are right about HDFS restart, also mentioned as step #2.

[#9]

thanks @Guilherme Braccialli - that makes sense.

[#10]

We ran into the same issue using Ambari 2.2.2.2.It stops at 35%.Says another instance is running and exiting. when you restart form Ambari or from the command line .Fix is :go to your name node run ps -aef|grep balancer kill -9 all the balancer process except the irq balancer (Linux native).Start the balancer from cmd line using # sudo -u hdfs hdfs balancer it works !!!

[#11]

@Kuldeep Kulkarni

I see the error in HDP 2.4.2, Ambari 2.2.2.0 version as well. HA is enabled. But as per the fix, it should be resolved in this version of ambari right... ?? Do you know why this error could be persisting ??