Running Accumulo on Fedora 25

Author  Christopher Tubbs and Mike Miller
Date   19 Dec 2016
Reviewer(s)   Keith Turner, Mike Walch

Apache Accumulo has been available in Fedora since F20. Recently, the Fedora packages have been updated to Accumulo version 1.6.6 and have made some improvements to the default configuration and launch scripts to provide a good out-of-box experience. This post will discuss the basic setup procedures for running Accumulo in the latest version, Fedora 25.

Prepare the system

WARNING: Before you start, be sure you’ve got plenty of free disk space. Otherwise, you could run into this bug or see other problems.

These instructions will assume you’re using Fedora 25, fully up-to-date (sudo dnf --refresh upgrade).

Install packages

Fedora provides a meta-package to install Accumulo and all of its dependencies. It’s a good idea to install the JDK, so you’ll have access to the jps command, and tuned for setting system performance tuning parameters from a profile. It’s also a good idea to ensure the optional hadoop native libraries are installed, and you have a good editor (replace vim with your preferred editor):

sudo dnf install accumulo java-1.8.0-openjdk-devel tuned vim hadoop-common-native

It is possible to install only a specific Accumulo service. For the single node setup, almost everything is needed. For the multi-node setup, it might make more sense to be selective about which you choose to install on each node (for example, to only install accumulo-tserver).

Set up tuned

(Optional) tuned can optimize your server settings, adjusting things like your vm.swappiness. To set up tuned, do:

sudo systemctl start tuned.service     # start service
sudo tuned-adm profile network-latency # pick a good profile
sudo tuned-adm active                  # verify the selected profile
sudo systemctl enable tuned.service    # auto-start on reboots

Set up ZooKeeper

You’ll need to set up ZooKeeper, regardless of whether you’ll be running a single node or many. So, let’s create its configuration file (the defaults are fine):

sudo cp /etc/zookeeper/zoo_sample.cfg /etc/zookeeper/zoo.cfg

Now, let’s start ZooKeeper (and set it to run on reboot):

sudo systemctl start zookeeper.service
sudo systemctl enable zookeeper.service

Note that the default port for ZooKeeper is 2181. Remember the hostname of the node where ZooKeeper is running, referred to as <zk-dns-name> later.

Running a single node

Configure Accumulo

To run on a single node, you don’t need to run HDFS. Accumulo can use the local filesystem as a volume instead. By default, it uses /tmp/accumulo. Let’s change that to something which will survive a reboot:

sudo vim /etc/accumulo/accumulo-site.xml

Change the value of the instance.volumes property from file:///tmp/accumulo to file:///var/tmp/accumulo in the configuration file (or another preferred location).

While you are editing the Accumulo configuration file, you should also change the default instance.secret from DEFAULT to something else. You can also change the credentials used by the tracer service now, too. If you use the root user, you’ll have to set its password to the same one you’ll use later when you initialize Accumulo. If you use another user name, you’ll have to create that user later.

Configure Hadoop client

Hadoop’s default local filesystem handler isn’t very good at ensuring files are written to disk when services are stopped. So, let’s use a better filesystem implementation for file:// locations. This implementation may not be as robust as a full HDFS instance, but it’s more reliable than the default. Even though you’re not going to be running HDFS, the Hadoop client code used in Accumulo can still be configured by modifying Hadoop’s configuration file:

sudo vim /etc/hadoop/core-site.xml

Add a new property:

  <property>
    <name>fs.file.impl</name>
    <value>org.apache.hadoop.fs.RawLocalFileSystem</value>
  </property>

Initialize Accumulo

Now, initialize Accumulo. You’ll need to do this as the accumulo user, because the Accumulo services run as the accumulo user. This user is created automatically by the RPMs if it doesn’t exist when the RPMs are installed. If you already have a user and/or group by this name, it will probably not be a problem, but be aware that this user will have permissions for the server configuration files. To initialize Accumulo as a specific user, use sudo -u:

sudo -u accumulo accumulo init

As expected, this command will fail if ZooKeeper is not running, or if the destination volume (file:///var/tmp/accumulo) already exists.

Start Accumulo services

Now that Accumulo is initialized, you can start its services:

sudo systemctl start accumulo-{master,tserver,gc,tracer,monitor}.service

Enable the commands to start at boot:

sudo systemctl enable accumulo-{master,tserver,gc,tracer,monitor}.service

Running multiple nodes

Amazon EC2 setup

For a multi-node setup, the authors tested these instructions with a Fedora 25 Cloud AMI on Amazon EC2 with the following characteristics:

  • us-east-1 availability zone
  • ami-e5757bf2 (latest in us-east-1 at time of writing)
  • HVM virtualization type
  • gp2 disk type
  • 64GB EBS root volume (no additional storage)
  • m4.large and m4.xlarge instance types (tested on both)
  • 3 nodes

For this setup, you should have a name service configured properly. For convenience, we used the EC2 provided internal DNS, with internal IP addresses. Make sure the nodes can communicate with each other using these names. If you’re using EC2, this means making sure they are in the same security group, and the security group has an inbound rule for “All traffic” with the source set to itself (sg-xxxxxxxx).

The default user is fedora for the Fedora Cloud AMIs. For the best experience, don’t forget to make sure they are fully up-to-date (sudo dnf --refresh upgrade).

Configure and run Hadoop

Configuring HDFS is the primary difference between the single and multi-node setup. For both Hadoop and Accumulo, you can edit the configuration files on one machine, and copy them to the others.

Pick a server to be the NameNode and identify its DNS name, (<namenode-dns-name>). Edit Hadoop’s configuration to set the default filesystem name to this location:

sudo vim /etc/hadoop/core-site.xml

Set the value for the property fs.default.name to hdfs://<namenode-dns-name>:8020.

Distribute copies of the changed configuration files to each node.

Now, format the NameNode. You’ll need to do this as the hdfs user on the NameNode instance:

sudo -u hdfs hdfs namenode -format

On the NameNode, start the NameNode service and enable it on reboot:

sudo systemctl start hadoop-namenode.service
sudo systemctl enable hadoop-namenode.service

On each DataNode, start the DataNode service:

sudo systemctl start hadoop-datanode.service
sudo systemctl enable hadoop-datanode.service

Configure and run Accumulo

Update Accumulo’s configuration to use this HDFS filesystem:

sudo vim /etc/accumulo/accumulo-site.xml

Change the value of the instance.volumes to hdfs://<namenode-dns-name>:8020/accumulo in the configuration file. Don’t forget to also change the default instance.secret and the trace user’s credentials, if necessary. Also, since you will have multiple nodes, you cannot use localhost:2181 for ZooKeeper, so set instance.zookeeper.host to <zk-dns-name>:2181.

Distribute copies of the changed configuration files to each node.

With HDFS now running, make sure Accumulo has permission to create its directory in HDFS, and initialize Accumulo:

sudo -u hdfs hdfs dfs -chmod 777 /
sudo -u accumulo accumulo init

After Accumulo has created its directory structure, you can change the permissions for the root back to what they were:

sudo -u hdfs hdfs dfs -chmod 755 /

Note: we only choose to do the above because this is a developer/testing environment. Temporarily changing ownership of HDFS is not recommended for the root of HDFS.

Now, you can start Accumulo.

On the NameNode, start all the Accumulo services and enable on reboot:

sudo systemctl start accumulo-{master,tserver,gc,tracer,monitor}.service
sudo systemctl enable accumulo-{master,tserver,gc,tracer,monitor}.service

On each DataNode, start just the tserver and enable it on reboot:

sudo systemctl start accumulo-tserver.service
sudo systemctl enable accumulo-tserver.service

Watching and using Accumulo

Run the shell

Run a shell as Accumulo’s root user (the instance name and root password are the ones you selected during the initialize step above:

accumulo shell -u root -zh <zk-dns-name>:2181 -zi <instanceName>

View the monitor pages

You should also be able to view the NameNode monitor page and the Accumulo monitor pages. If you are running this in EC2, you can view these over an SSH tunnel using the NameNode’s public IP address. If you didn’t give this node a public IP address, you can allocate one in EC2 and associate it with this node:

ssh -L50070:localhost:50070 -L50095:localhost:50095 <user>@<host>

Replace <user> with your username (probably fedora if using the Fedora AMI), and <host> with the public IP or hostname for your EC2 instance. Now, in your local browser, you should be able to navigate to these addresses in your localhost: Hadoop monitor (http://localhost:50070) and Accumulo monitor (http://localhost:50095).

Debugging commands

Check the status of a service:

sudo systemctl status <ServiceName>.service

Check running Java processes:

sudo jps -ml

Check the system logs for a specific service within the last 10 minutes:

sudo journalctl -u <ServiceName> --since '10 minutes ago'

Check listening ports:

sudo netstat -tlnp

Check DNS name for a given IP address:

getent hosts <ipaddress> # OR
hostname -A

Perform forward and reverse DNS lookups:

sudo dnf install bind-utils
dig +short <hostname>     # forward DNS lookup
dig +short -x <ipaddress> # reverse DNS lookup

Find the instance ID for your instance name:

zkCli.sh -server <host>:2181     # replace <host> with your ZooKeeper server DNS name
> get /accumulo/instances/<name> # replace <name> with your instance name
> quit

If the NameNode is listening on the loopback address, you’ll probably need to restart the service manually, as well as any Accumulo services which failed. This is a known issue with Hadoop:

sudo systemctl restart hadoop-namenode.service

Some helpful rpm commands:

rpm -q -i <installed-package-name>              # to see info about an installed package
rpm -q -i -p <rpm-file-name>                    # to see info about an rpm file
rpm -q --provides <installed-package-name>      # see what a package provides
rpm -q --requires <installed-package-name>      # see what a package requires
rpm -q -l <installed-package-name>              # list package files
rpm -q --whatprovides <file>                    # find rpm which owns <file>
rpm -q --whatrequires 'mvn(groupId:artifactId)' # find rpm which requires maven coords

Helping out

Feel free to get involved with the Fedora or Fedora EPEL (for RHEL/CentOS users) packaging. Contact the Fedora maintainers (user at fedoraproject dot org) for the Accumulo packages to see how you can help patching bugs, adapting the upstream packages to the Fedora packaging standards, testing updates, maintaining dependency packages, and more.

View all posts in the news archive