Apache Accumulo Testing Suite

The Apache Accumulo testing suite contains applications that test and verify the correctness of Accumulo.

Installation

In order to run the Apache Accumulo testing suite, you will need Java 8 and Maven installed on your machine as well as an Accumulo instance to use for testing.

1. First clone this repository.

git clone [email protected]:apache/accumulo-testing.git
cd accumulo-testing

2. All configuration files for the test suite are in conf/. Only the accumulo-testing.properties configuration file needs to be edited as all other configuration files are optional. In accumulo-testing.properites, review the properties with test.common.* prefix as these are used by all tests.

cd conf/
vim accumulo-testing.properties

Run tests locally

Tests are run using the following scripts in bin/:

• cingest - Runs continuous ingest tests
• rwalk - Runs random walk tests
• performance - Runs performance test
• agitator - Runs agitator
• gcs - Runs garbage collection simulation
• monitor - Runs availability monitor probe

Run the scripts without arguments to view usage.

Run tests in Docker

While test scripts can be run from a single machine, they will put more stress if they are run from multiple machines. The easiest way to do this is using Docker. However, only the tests below can be run in Docker:

• cingest - All applications can be run except verify & moru which launch a MapReduce job.
• rwalk - All modules can be run.
• monitor - All modules can be run.

1. To create the accumulo-testing docker image, make sure the following files exist in your clone:

   • conf/accumulo-client.properties - Configure this file from your Accumulo install
   • conf/accumulo-testing.properties - Configure this file for testing
   • target/accumulo-testing-2.1.0-SNAPSHOT-shaded.jar - Can be created using ./bin/build

   Run the following command to create the image. HADOOP_HOME should be where Hadoop is installed on your cluster. HADOOP_USER_NAME should match the user running Hadoop on your cluster.

   docker build --build-arg HADOOP_HOME=$HADOOP_HOME --build-arg HADOOP_USER_NAME=`whoami` -t accumulo-testing .

2. The accumulo-testing image can run a single command:

   docker run --network="host" accumulo-testing cingest createtable

3. Multiple containers can also be run (if you have Docker Swarm enabled):

   # the following can be used to get the image on all nodes if you do not have a registry.
   for HOST in node1 node2 node3; do
     docker save accumulo-testing | ssh -C $HOST docker load &
   done
   
   docker service create --network="host" --replicas 2 --name ci accumulo-testing cingest ingest

Random walk test

The random walk test generates client behavior on an Apache Accumulo instance by randomly walking a graph of client operations.

Before running random walk, review the test.common.* properties in accumulo-testing.properties file. A test module must also be specified. See the modules directory for a list of available ones.

The command below will start a single random walker in a local process using the Image.xml module.

./bin/rwalk Image.xml

Continuous Ingest & Query

The Continuous Ingest test runs many ingest clients that continually create linked lists of data in Accumulo. During ingest, query applications can be run to continuously walk and verify the linked lists and put a query load on Accumulo. At some point, the ingest clients are stopped and a MapReduce job is run to ensure that there are no holes in any linked list.

The nodes in the linked list are random. This causes each linked list to spread across the table. Therefore, if one part of a table loses data, then it will be detected by references in another part of table.

Before running any of the Continuous Ingest applications, make sure that the accumulo-testing.properties file exists in conf/ and review all properties with the test.ci.* prefix.

First, run the command below to create an Accumulo table for the continuous ingest tests. The name of the table is set by the property test.ci.common.accumulo.table (its value defaults to ci) in the file accumulo-testing.properties:

./bin/cingest createtable {-o test.<prop>=<value>}

The continuous ingest tests have several applications that start a local application which will run continuously until you stop using ctrl-c:

./bin/cingest <application> {-o test.<prop>=<value>}

Below is a list of available continuous ingest applications. You should run the ingest application first to add data to your table.

• ingest - Inserts data into Accumulo that will form a random graph.
• walk - Randomly walks the graph created by ingest application using scanner. Each walker produces detailed statistics on query/scan times.
• batchwalk - Randomly walks the graph created by ingest using a batch scanner.
• scan - Scans the graph
• verify - Runs a MapReduce job that verifies all data created by continuous ingest. Before running, review all test.ci.verify.* properties. Do not run ingest while running this command as it will cause erroneous reporting of UNDEFINED nodes. Each entry, except for the first batch of entries, inserted by continuous ingest references a previously flushed entry. Since we are referencing flushed entries, they should always exist. The MapReduce job checks that all referenced entries exist. If it finds any that do not exist it will increment the UNDEFINED counter and emit the referenced but undefined node. The MapReduce job produces two other counts: REFERENCED and UNREFERENCED. It is expected that these two counts are non-zero. REFERENCED counts nodes that are defined and referenced. UNREFERENCED counts nodes that defined and unreferenced, these are the latest nodes inserted.
• bulk - Runs a MapReduce job that generates data for bulk import. See bulk-test.md.
• moru - Runs a MapReduce job that stresses Accumulo by reading and writing the continuous ingest table. This MapReduce job will write out an entry for every entry in the table (except for ones created by the MapReduce job itself). Stop ingest before running this MapReduce job. Do not run more than one instance of this MapReduce job concurrently against a table.

Check out ingest-test.md for pointers on running a long-running ingest and verification test.

Garbage Collection Simulator

See gcs.md.

Agitator

The agitator will periodically kill the Accumulo manager, tablet server, and Hadoop data node processes on random nodes. Before running the agitator you should create accumulo-testing-env.sh in conf/ and review all the agitator settings. The command below will start the agitator:

./bin/agitator start

Running this script as root will properly start processes as the user you configured in env.sh (AGTR_HDFS_USER for the data node and AGTR_ACCUMULO_USER for Accumulo processes). If you run it as yourself and the AGTR_HDFS_USER and AGTR_ACCUMULO_USER values are the same as your user, the agitator will not change users. In the case where you run the agitator as a non-privileged user which isn't the same as AGTR_HDFS_USER or AGTR_ACCUMULO_USER, the agitator will attempt to sudo to these users, which relies on correct configuration of sudo. Also, be sure that your AGTR_HDFS_USER has password-less ssh configured.

Run the command below stop the agitator:

./bin/agitator stop

Performance Test

To run performance test a cluster-control.sh script is needed to assist with starting, stopping, wiping, and configuring an Accumulo instance. This script should define the following functions.

function get_hadoop_client {
  # TODO return hadoop client libs in a form suitable for appending to a classpath
}

function get_version {
  case $1 in
    ACCUMULO)
      # TODO echo accumulo version
      ;;
    HADOOP)
      # TODO echo hadoop version
      ;;
    ZOOKEEPER)
      # TODO echo zookeeper version
      ;;
    *)
      return 1
  esac
}

function start_cluster {
  # TODO start Hadoop and Zookeeper if needed
}

function setup_accumulo {
  # TODO kill any running Accumulo instance
  # TODO setup a fresh install of Accumulo w/o starting it
}

function get_config_file {
  local file_to_get=$1
  local dest_dir=$2
  # TODO copy $file_to_get from Accumulo conf dir to $dest_dir
}

function put_config_file {
  local config_file=$1
  # TODO copy $config_file to Accumulo conf dir
}

function put_server_code {
  local jar_file=$1
  # TODO add $jar_file to Accumulo's server side classpath. Could put it in $ACCUMULO_HOME/lib/ext
}

function start_accumulo {
  # TODO start accumulo
}

function stop_cluster {
  # TODO kill Accumulo, Hadoop, and Zookeeper
}

An example script for Uno is provided. To use this, do the following and set UNO_HOME after copying.

cp conf/cluster-control.sh.uno conf/cluster-control.sh

After the cluster control script is set up, the following will run performance test and produce JSON result files in the provided output directory.

./bin/performance run <output dir>

The example above will run all performance tests in order. To run a single test, a filter can be applied. The example below will run just the DurabilityWriteSpeedPT.

./bin/performance run <output dir> DurabilityWriteSpeedPT

Some performance tests alter the system properties of the cluster it is trying to test on. These may require fine-tuning in order to work on some hardware.

There are some utilities for working with the JSON result files, run the performance script with no options to see them.

Availability Monitor

Monitor class aims at verifying availability of overall accumulo cluster by continually doing scans of random values across various tablet servers and capturing timing information related to how long such scans take.

Automated Cluster Testing

See the readme.md.