Each node contributes to the cluster designed in the previous step, so each must be able to run MapR and Hadoop software.
Red Hat, CentOS, SUSE, or Ubuntu
8 GB minimum, more in production
Raw, unformatted drives and partitions
Hostname, reaches all other nodes
Common users across all nodes; passwordless ssh (optional)
Must run Java
NTP, Syslog, PAM
Use the following sections as a checklist to make each candidate node suitable for its assigned roles. Once each node has been prepared or disqualified, proceed to Step 3, Installing MapR Software.
2.1 CPU and Operating System
a. Processor is 64-bit
To determine the processor type, run
If the output includes "x86_64," the processor is 64-bit. If it includes "i386," "i486," "i586," or "i686," it is a 32-bit processor, which is not supported by MapR software.
If the results are "unknown," or none of the above, try one of these alternative commands.
cpuinfo file, the flag 'lm' (for "long-mode") indicates a 64-bit processor.
b. Operating System is supported
Run the following command to determine the name and version of the installed operating system. (If the lsb_release command reports "No LSB modules are available," this is not a problem.)
The operating system must be one of the following:
RedHat Enterprise Linux (RHEL) or Community Enterprise Linux (CentOS)
6.1 or later
SUSE Enterprise Linux Server
11 SP2 or later
12.04 or later
lsb_release command is not found, try one of the following alternatives.
If you determine that the node is running an older version of a supported OS, upgrade to at least a supported version and test the upgrade before proceeding. If you find a different Linux distribution, such as Fedora or Gentoo, the node must be reformatted and a supported distro installed.
2.2 Memory and Disk Space
a. Minimum Memory
A minimum of 8 GB total memory is required on a node. MapR recommends at least 16 GB for a production environment, and typical MapR production nodes have 32 GB or more.
free -g to display total and available memory in gigabytes.
free command is not found, there are many alternatives:
grep MemTotal: /proc/meminfo,
vmstat -s -SM,
top, or various GUI system information tools.
MapR does not recommend using the numad service, since it has not been tested and validated with MapR. Using numad can cause artificial memory constraints to be set which can lead to performance degradation under load. To disable numad:
- Stop the service by issuing the command
service numad stop.
- Set the numad service not to start on reboot:
chkconfig numad off
MapR does not recommend using overcommit because it may lead to the kernel memory manager killing processes to free memory, resulting in killed MapR processes and system instability. Set
vm.overcommit_memory to 0:
Edit the file
/etc/sysctl.confand add the following line:
Save the file and run:
MapR manages raw, unformatted devices directly to optimize performance and offer high availability. For data nodes, allocate at least 3 unmounted physical drives or partitions for MapR storage. MapR uses disk spindles in parallel for faster read/write bandwidth and therefore groups disks into sets of three.
Do not use RAID or Logical Volume Management with disks that will be added to MapR. While MapR supports these technologies, using them incurs additional setup overhead and can affect your cluster's performance. Due to the possible formatting requirements that are associated with changes to the drive settings, configure the drive settings prior to installing MapR.
If you have a RAID controller, configure it to run in HBA mode. For LSI MegaRAID controllers that do not support HBA, configure the following drive group settings for optimal performance:
|Property (The actual name depends on the version)||Recommended Setting|
|Cache Policy or I/O Policy||Cached IO or Cached|
|Read Policy||Always Read Ahead or Read Ahead|
|Disk Cache Policy or Drive Cache||Disabled|
Minimum Disk Space
OS Partition. Provide at least 10 GB of free disk space on the operating system partition.
MapR-FS. Provide at least 8 GB of free disk space for MapR-FS.
Disk. Provide 10 GB of free disk space in the
/tmp directory and 128 GB of free disk space in the
/opt directory. Services, such as JobTracker and TaskTracker, use the
/tmp directory. Files, such as logs and cores, use the
Swap space. Provide sufficient swap space for stability, 10% more than the node's physical memory, but not less than 24 GB and not more than 128 GB.
ZooKeeper. On ZooKeeper nodes, dedicate a partition, if practicable, for the
/opt/mapr/zkdata directory to avoid other processes filling that partition with writes and to reduce the possibility of errors due to a full
/opt/mapr/zkdata directory. This directory is used to store snapshots that are up to 64 MB. Since the four most recent snapshots are retained, reserve at least 500 MB for this partition. Do not share the physical disk where
/opt/mapr/zkdata resides with any MapR File System data partitions to avoid I/O conflicts that might lead to ZooKeeper service failures.
Each node in the cluster must have a unique hostname, resolvable forward and backward with every other node with both normal and reverse DNS name lookup.
hostname -f to check the node's hostname. For example:
hostname -f returns a name, run
getent hosts `hostname` to return the node's IP address and fully-qualified domain name (FQDN).
To troubleshoot hostname problems, edit the
/etc/hosts file as root. A simple
/etc/hosts might contain:
A common problem is an incorrect loopback entry (127.0.x.x) that prevents the IP address from being assigned to the hostname. For example, on Ubuntu, the default
/etc/hosts file might contain:
A loopback (
127.0.x.x) entry with the node's hostname will confuse the installer and other programs. Edit the
/etc/hosts file and delete any entries that associate the hostname with a loopback IP. Only associate the hostname with the actual IP address.
ping command to verify that each node can reach the others using each node's hostname. For more information, see the hosts(5) man page.
b. Common Users
A user that accesses the cluster must have the same credentials and user ID (uid) on each node in the cluster. Every person or department that runs MapR jobs must have an account and must also belong to a common group ID (gid). The uid for each user, and the gid for each group, must be consistent across all nodes.
A 'mapr' user must exist. The 'mapr' user has full privileges to administer the cluster. If you create the 'mapr' user before you install MapR, you can test for connectivity issues. If you do not create the 'mapr' user, installing MapR automatically creates the user for you. The 'mapr' user ID is automatically created on each node if you do not use a directory service, such as LDAP.
To create a group, add a user to the group, or create the 'mapr' user, run the following command as root substituting a uid for m and a gid for n. (The error "cannot lock /etc/passwd" suggests that the command was not run as root.)
$ groupadd -g 5000 mapr
$ useradd -g 5000 -u 5000 mapr
To verify that the users or groups were created,
su mapr. Verify that a home directory was created (usually
/home/mapr) and that the users or groups have read-write access to it. The users or groups must have write access to the /tmp directory, or the warden will fail to start services.
c. Optional: Passwordless ssh
If you plan to use the scripted rolling upgrade procedure to upgrade the cluster in the future, it is very helpful for the common user to be able to ssh from each webserver node to any other node without providing a password. Otherwise, passwordless ssh between nodes is optional because MapR will run without it.
Setting up passwordless ssh is straightforward. On each webserver node, generate a key pair and append the key to an authorization file. Then copy this authorization file to each node, so that every node is available from the webserver node.
su mapr (if you are not already logged in as mapr)
ssh-keygen -t rsa -P '' -f ~/
ssh-keygen command creates
filename, containing the private key, and
.pub, containing the public key. For convenience, you may want to name the file for the hostname of the node. For example, on the node with hostname "node10.10.1.1,"
ssh-keygen -t rsa -P '' -f ~/node10.10.1.1
In this example, append the file
/home/mapr/node10.10.1.1.pub to the
Append each webserver node's public key to a single file, using a command like
.pub >> authorized_keys. (The key file is simple text, so you can append the file in several ways, including a text editor.) When every webserver node's empty passphrase public key has been generated, and the public key file has been appended to the master "authorized_keys" file, copy this master keys file to each node as
~/.ssh/authorized_keys, where ~ refers to the mapr user's home directory (typically
MapR services require the Java runtime environment (JRE) and Java Development Kit (JDK).
java -version. Verify that one of these versions is installed on the node:
- Sun Java JDK 1.7 or 1.8
- OpenJDK 1.7 or 1.8
java command is not found, download and install Oracle/Sun Java or use a package manager to install OpenJDK. Obtain the Oracle/Sun Java Runtime Environment (JRE), Standard Edition (Java SE), available at Oracle's Java SE website. Find Java SE 7 in the archive of previous versions.
Use a package manager, such as yum (RedHat or CentOS), apt-get (Ubuntu) or zypper (SuSE) to install or update OpenJDK on the node. The command will be something like one of these:
# yum install java-1.7.0-openjdk-devel
# apt-get install openjdk-7-jdk
# zypper install openjdk-7-jdk
The MapR Metrics service requires access to a MySQL server running version 5.1 or later. MySQL does not have to be installed on a node in the cluster, but it must be on the same network as the cluster. If you do not plan to use MapR Metrics, MySQL is not required.
a. Network Time
To keep all cluster nodes time-synchronized, MapR requires software such as a Network Time Protocol (NTP) server to be configured and running on every node. If server clocks in the cluster drift out of sync, serious problems will occur with HBase and other MapR services. MapR raises a Time Skew alarm on any out-of-sync nodes. See http://www.ntp.org/ for more information about obtaining and installing NTP.
Advanced: Installing an internal NTP server keeps your cluster synchronized even when an outside NTP server is inaccessible.
Syslog must be enabled on each node to preserve logs regarding killed processes or failed jobs. Modern versions such as syslog-ng and rsyslog are possible, making it more difficult to be sure that a syslog daemon is present. One of the following commands should suffice:
c. Default umask
Ensure that the default umask for the root user is set to 0022 on all mapr nodes in the cluster. The umask setting is changed in the /etc/profile file, or in the .cshrc or .login file. The root user must have a 0022 umask because the MapR admin user requires access to all files and directories under the /opt/mapr directory, even those initially created by root services.
ulimit is a command that sets limits on the user's access to system-wide resources. Specifically, it provides control over the resources available to the shell and to processes started by it.
The mapr-warden script uses the
ulimit command to set the maximum number of file descriptors (
nofile) and processes (
nproc) to 64000. Higher values are unlikely to result in an appreciable performance gain. Lower values, such as the default value of 1024, are likely to result in task failures.
Depending on your environment, you might want to set limits manually rather than relying on Warden to set them automatically using
ulimit. The following examples show how to do this, using the recommended value of 64000.
/etc/security/limits.confand add the following line:
/etc/security/limits.d/90-nproc.confand add the following line:
Check that the
/etc/pam.d/sufile contains the following settings:
/etc/security/limits.confand add the following lines:
/etc/pam.d/suand uncomment the following line:
ulimit to verify settings:
- Reboot the system.
Run the following command as the MapR user (not root) at a command line:
The command should report
Nodes that will run the MapR Control System (the
mapr-webserver service) can take advantage of Pluggable Authentication Modules (PAM) if found. Configuration files in
/etc/pam.d/ directory are typically provided for each standard Linux command. MapR can use, but does not require, its own profile.
For more detail about configuring PAM, see PAM Configuration.
f. Security - SELinux, AppArmor
SELinux (or the equivalent on other operating systems) must be disabled during the install procedure. If the MapR services run as a non-root user, SELinux can be enabled after installation and while the cluster is running.
g. TCP Retries
On each node, set the number of TCP retries to 5 so that MapR can detect unreachable nodes with less latency.
Edit the file
/etc/sysctl.confand add the following line:
Save the file and run:
Disable the stock Linux NFS server on nodes that will run the MapR NFS server.
Enabling iptables on a node may close ports that are used by MapR. If you enable iptables, make sure that required ports remain open. Check your current IP table rules with the following command:
j. Transparent Huge Pages (THP)
For data-intensive workloads, MapR recommends disabling the Transparent Huge Pages (THP) feature in the Linux kernel.
Some users find tools like Puppet or Chef useful to configure each node in a cluster. Make sure, however, that any configuration tool does not reset changes made when MapR packages are later installed. Specifically, do not let automated configuration tools overwrite changes to the following files:
Each prospective node in the cluster must be checked against the requirements presented here. Failure to ensure that each node is suitable for use generally leads to hard-to-resolve problems with installing Hadoop.
After each node has been shown to meet the requirements and has been prepared, you are ready to Install MapR components.