Under‐replicated block exceptions or cluster failure occurs on small clusters

You can troubleshoot the under-replication warning by editing the dfs.replication.max and the table.file.replication properties.

By default, Operational Database powered by Apache Accumulo (OpDB) attempts to use a replication factor of five for the HDFS files that make up the metadata table. OpDB ignores the table.file.replication setting. This causes under‐replication warnings until you correct the number of nodes or until you manually adjust the replication setting on that table.

Normally, Cloudera Manager does not set a maximum replication factor. If it is set, OpDB uses the table.file.replication setting only on normal user tables to that maximum.

However, if the dfs.replication.max setting for HDFS is adjusted to match the number of cluster nodes, attempts by OpDB to create new files for its internal tables will fail on clusters with fewer than five datanodes.

  1. Edit the dfs.replication.max setting for HDFS to be more or equal to 5.
  2. Adjust replication on the metadata and root tables to be less than or equal to the number of datanodes.
  3. Lower the dfs.replication.max setting value.
    For example, to adjust the replication in the Accumulo shell to 3:
    root@accumulo> config -t accumulo.metadata -s table.file.replication=3 
    root@accumulo> config -t accumulo.root -s table.file.replication=3