The randomwalk framework needs to be configured for your Accumulo instance by doing the following steps:
Make sure you have both ACCUMULO_HOME and HADOOP_HOME set in your $ACCUMULO_CONF_DIR/accumulo-env.sh.
Create ‘randomwalk.conf’ file in the conf directory containing settings needed by walkers to connect to Accumulo.
Create a ‘walkers’ file in the conf directory containing the hostnames of the machines where you want random walkers to run.
Create a ‘logger.xml’ file in the conf directory from logger.xml.example.
The command below starts random walkers on all machines listed in ‘walkers’. The argument Image.xml indicates the module to use (which is located at conf/modules/Image.xml):
$ ./bin/start-all.sh Image.xml
All modules must be in conf/modules and can be referenced without this prefix. For example, a module located at conf/modules/foo/bar.xml is started as the following:
$ ./bin/start-all.sh foo/bar.xml
This command will load all configuration in the conf directory to HDFS and start identical random walkers on each node. These random walkers will download the current configuration from HDFS and place them in the tmp/ directory.
Random walkers will drop their logs in the logs/ directory. If you are running multiple walkers and want ERROR/WARNs dropped to an NFS-hosted log, please set NFS_LOGPATH to a NFS-mounted directory and uncomment the NFS appender in logger.xml.
You can kill all walkers on the machines listed in the ‘walkers’ file using the following command:
$ ./bin/kill-all.sh
The user accounts for walkers that run the Concurrent.xml module must have password-less SSH access to the entire Accumulo cluster, so that they may run $ACCUMULO_HOME/bin/start-all.sh. Note that this is not the same script as the one that starts random walkers; it is the script to start up an Accumulo cluster. You can test that access is in place by running $ACCUMULO_HOME/bin/start-all.sh from the command line of each walker account.
The above access is also needed for any modules that include Concurrent.xml, e.g., ShortClean.xml, LongClean.xml.
Copies configuration in conf/ to HDFS:
$ copy-config.sh
Copies configuration from HDFS into tmp/ and starts only one local random walker.
$ start-local.sh All.xml
Stops all local random walkers:
$ pkill -f randomwalk.Framework
If you are running randomwalk tests while exercising Hadoop's high availability (HA) failover capabilities, you should use Hadoop version 2.1.0 or later. Failover scenarios are more likely to cause randomwalk test failures under earlier Hadoop versions. See the following issue reports for more details.