There will not be any operational problems such as crashes or errors when running a HDFS balancer on a cluster with HBase running, but there can potentially be a performance impact depending on what the balancer decides to move based on its space thresholds. The performance impact would come from loss of locality - the RegionServers' required HFiles may find their blocks to be remote, so a slightly higher network usage can be observed until the next major compaction rewrites a block replica locally. If you'd like to narrow down the time-frame of impact, you can run the HDFS balancer with the desired balancing threshold, and then once it is complete, immediately follow up with a major compaction command on your latency-sensitive HBase tables. ... View more

HBase authz will not consult NameNode for groups but rather will check it local to each serving RegionServer handling the request. Ensure your Linux user and groups are consistent across _all_ cluster hosts for a predicable result with any authorization feature. ... View more

The documentation of CM metrics for HBase Master states the below: """ The number of times the balancer was invoked outside a balance cluster operation. The balancer might be invoked when a new table is created, for example, to place the newly created regions. """ - https://www.cloudera.com/documentation/enterprise/latest/topics/cm_metrics_master.html If you want more granularity on when this metric receives an increment, look for the method calls named 'incrMiscInvocations' inside the balancer base class within HBase Master: https://github.com/cloudera/hbase/blob/cdh5.15.0-release/hbase-server/src/main/java/org/apache/hadoop/hbase/master/balancer/BaseLoadBalancer.java -- The _across_masters suffix just averages the value across all HBase Masters running in the cluster. The _across_roletype generally applies the same way to all other role type metrics in CM. You can also query them without the across suffixes. The total_ prefix sums instead of averaging the value, across all relevant role types for your chosen metric. The rate bit basically appears for all counter style metrics. It represents the change over time (dx/dt). You can perform an integral(…) over any rate-named metrics to access actual value growth. These are tsquery specifics documented further at https://www.cloudera.com/documentation/enterprise/latest/topics/cm_dg_metric_aggregation.html#cmug_topic_11_8 ... View more

What version(s) of JDK/JRE are installed on the host that runs your NFS Gateway? Is it consistent with the other hosts? CDH/CM requires recent version(s) of Oracle JDK version 1.7 or version 1.8 to run: https://www.cloudera.com/documentation/enterprise/release-notes/topics/rn_consolidated_pcm.html#pcm_jdk and it is recommended to not keep multiple different version(s) of Java JRE/JDK installed. ... View more