Hi Dima,
This really comes down to the Namenode's FQDN being renamed, not the
rest of the Accumulo nodes.
Accumulo stores full URI's to files (e.g.
hdfs://nn:8020/accumulo/../foo.rf) which is why this is an issue.
Suddenly, Accumulo can't find any of the files in HDFS.
You can use instance.volumes.replacements in accumulo-site.xml to handle
the rename automatically.
http://accumulo.apache.org/1.7/accumulo_user_manual#_migrating_accumulo_from_non_ha_namenode_to_ha_namenode
has a concrete example of this in practice (moving from a single NN to a
HA nameservice).
- Josh
Dima Spivak wrote:
Hi all,
I recently had a situation where I had to change the domain of the
machines running an Accumulo cluster, and learned the hard way that
Accumulo doesn't like that. In particular, I began getting
UnknownHostExceptions on the master with messages complaining about not
being able to process "table state for store Root Table."
Anyone happen to know of a tool or even just some commands I'd need to
run to tell the cluster that the hostnames it had before are different
now and to stop being so unhappy? :)
-Dima
