This document describes how to install, configure and manage non-trivial Hadoop clusters ranging from a few nodes to extremely large clusters with thousands of nodes.
To play with Hadoop, you may first want to install it on a single machine (see Single Node Setup).
Download a stable version of Hadoop from Apache mirrors.
Installing a Hadoop cluster typically involves unpacking the software on all the machines in the cluster or installing RPMs.
Typically one machine in the cluster is designated as the NameNode and another machine the as ResourceManager, exclusively. These are the masters.
The rest of the machines in the cluster act as both DataNode and NodeManager. These are the slaves.
The following sections describe how to configure a Hadoop cluster.
Hadoop configuration is driven by two types of important configuration files:
Additionally, you can control the Hadoop scripts found in the bin/ directory of the distribution, by setting site-specific values via the conf/hadoop-env.sh and yarn-env.sh.
To configure the Hadoop cluster you will need to configure the environment in which the Hadoop daemons execute as well as the configuration parameters for the Hadoop daemons.
The Hadoop daemons are NameNode/DataNode and ResourceManager/NodeManager.
Administrators should use the conf/hadoop-env.sh and conf/yarn-env.sh script to do site-specific customization of the Hadoop daemons' process environment.
At the very least you should specify the JAVA_HOME so that it is correctly defined on each remote node.
In most cases you should also specify HADOOP_PID_DIR and HADOOP_SECURE_DN_PID_DIR to point to directories that can only be written to by the users that are going to run the hadoop daemons. Otherwise there is the potential for a symlink attack.
Administrators can configure individual daemons using the configuration options shown below in the table:
| Daemon | Environment Variable |
|---|---|
| NameNode | HADOOP_NAMENODE_OPTS |
| DataNode | HADOOP_DATANODE_OPTS |
| Secondary NameNode | HADOOP_SECONDARYNAMENODE_OPTS |
| ResourceManager | YARN_RESOURCEMANAGER_OPTS |
| NodeManager | YARN_NODEMANAGER_OPTS |
| WebAppProxy | YARN_PROXYSERVER_OPTS |
| Map Reduce Job History Server | HADOOP_JOB_HISTORYSERVER_OPTS |
For example, To configure Namenode to use parallelGC, the following statement should be added in hadoop-env.sh :
export HADOOP_NAMENODE_OPTS="-XX:+UseParallelGC ${HADOOP_NAMENODE_OPTS}"Other useful configuration parameters that you can customize include:
| Daemon | Environment Variable |
|---|---|
| ResourceManager | YARN_RESOURCEMANAGER_HEAPSIZE |
| NodeManager | YARN_NODEMANAGER_HEAPSIZE |
| WebAppProxy | YARN_PROXYSERVER_HEAPSIZE |
| Map Reduce Job History Server | HADOOP_JOB_HISTORYSERVER_HEAPSIZE |
This section deals with important parameters to be specified in the given configuration files:
| Parameter | Value | Notes |
|---|---|---|
| fs.defaultFS | NameNode URI | hdfs://host:port/ |
| io.file.buffer.size | 131072 | Size of read/write buffer used in SequenceFiles. |
| Parameter | Value | Notes |
|---|---|---|
| dfs.namenode.name.dir | Path on the local filesystem where the NameNode stores the namespace and transactions logs persistently. | If this is a comma-delimited list of directories then the name table is replicated in all of the directories, for redundancy. |
| dfs.namenode.hosts / dfs.namenode.hosts.exclude | List of permitted/excluded DataNodes. | If necessary, use these files to control the list of allowable datanodes. |
| dfs.blocksize | 268435456 | HDFS blocksize of 256MB for large file-systems. |
| dfs.namenode.handler.count | 100 | More NameNode server threads to handle RPCs from large number of DataNodes. |
| Parameter | Value | Notes |
|---|---|---|
| dfs.datanode.data.dir | Comma separated list of paths on the local filesystem of a DataNode where it should store its blocks. | If this is a comma-delimited list of directories, then data will be stored in all named directories, typically on different devices. |
| Parameter | Value | Notes |
|---|---|---|
| yarn.acl.enable | true / false | Enable ACLs? Defaults to false. |
| yarn.admin.acl | Admin ACL | ACL to set admins on the cluster. ACLs are of for comma-separated-usersspacecomma-separated-groups. Defaults to special value of * which means anyone. Special value of just space means no one has access. |
| yarn.log-aggregation-enable | false | Configuration to enable or disable log aggregation |
| Parameter | Value | Notes |
|---|---|---|
| yarn.resourcemanager.address | ResourceManager host:port for clients to submit jobs. | host:port |
| yarn.resourcemanager.scheduler.address | ResourceManager host:port for ApplicationMasters to talk to Scheduler to obtain resources. | host:port |
| yarn.resourcemanager.resource-tracker.address | ResourceManager host:port for NodeManagers. | host:port |
| yarn.resourcemanager.admin.address | ResourceManager host:port for administrative commands. | host:port |
| yarn.resourcemanager.webapp.address | ResourceManager web-ui host:port. | host:port |
| yarn.resourcemanager.scheduler.class | ResourceManager Scheduler class. | CapacityScheduler (recommended), FairScheduler (also recommended), or FifoScheduler |
| yarn.scheduler.minimum-allocation-mb | Minimum limit of memory to allocate to each container request at the Resource Manager. | In MBs |
| yarn.scheduler.maximum-allocation-mb | Maximum limit of memory to allocate to each container request at the Resource Manager. | In MBs |
| yarn.resourcemanager.nodes.include-path / yarn.resourcemanager.nodes.exclude-path | List of permitted/excluded NodeManagers. | If necessary, use these files to control the list of allowable NodeManagers. |
| Parameter | Value | Notes |
|---|---|---|
| yarn.nodemanager.resource.memory-mb | Resource i.e. available physical memory, in MB, for given NodeManager | Defines total available resources on the NodeManager to be made available to running containers |
| yarn.nodemanager.vmem-pmem-ratio | Maximum ratio by which virtual memory usage of tasks may exceed physical memory | The virtual memory usage of each task may exceed its physical memory limit by this ratio. The total amount of virtual memory used by tasks on the NodeManager may exceed its physical memory usage by this ratio. |
| yarn.nodemanager.local-dirs | Comma-separated list of paths on the local filesystem where intermediate data is written. | Multiple paths help spread disk i/o. |
| yarn.nodemanager.log-dirs | Comma-separated list of paths on the local filesystem where logs are written. | Multiple paths help spread disk i/o. |
| yarn.nodemanager.log.retain-seconds | 10800 | Default time (in seconds) to retain log files on the NodeManager Only applicable if log-aggregation is disabled. |
| yarn.nodemanager.remote-app-log-dir | /logs | HDFS directory where the application logs are moved on application completion. Need to set appropriate permissions. Only applicable if log-aggregation is enabled. |
| yarn.nodemanager.remote-app-log-dir-suffix | logs | Suffix appended to the remote log dir. Logs will be aggregated to ${yarn.nodemanager.remote-app-log-dir}/${user}/${thisParam} Only applicable if log-aggregation is enabled. |
| yarn.nodemanager.aux-services | mapreduce_shuffle | Shuffle service that needs to be set for Map Reduce applications. |
| Parameter | Value | Notes |
|---|---|---|
| yarn.log-aggregation.retain-seconds | -1 | How long to keep aggregation logs before deleting them. -1 disables. Be careful, set this too small and you will spam the name node. |
| yarn.log-aggregation.retain-check-interval-seconds | -1 | Time between checks for aggregated log retention. If set to 0 or a negative value then the value is computed as one-tenth of the aggregated log retention time. Be careful, set this too small and you will spam the name node. |
| Parameter | Value | Notes |
|---|---|---|
| mapreduce.framework.name | yarn | Execution framework set to Hadoop YARN. |
| mapreduce.map.memory.mb | 1536 | Larger resource limit for maps. |
| mapreduce.map.java.opts | -Xmx1024M | Larger heap-size for child jvms of maps. |
| mapreduce.reduce.memory.mb | 3072 | Larger resource limit for reduces. |
| mapreduce.reduce.java.opts | -Xmx2560M | Larger heap-size for child jvms of reduces. |
| mapreduce.task.io.sort.mb | 512 | Higher memory-limit while sorting data for efficiency. |
| mapreduce.task.io.sort.factor | 100 | More streams merged at once while sorting files. |
| mapreduce.reduce.shuffle.parallelcopies | 50 | Higher number of parallel copies run by reduces to fetch outputs from very large number of maps. |
| Parameter | Value | Notes |
|---|---|---|
| mapreduce.jobhistory.address | MapReduce JobHistory Server host:port | Default port is 10020. |
| mapreduce.jobhistory.webapp.address | MapReduce JobHistory Server Web UI host:port | Default port is 19888. |
| mapreduce.jobhistory.intermediate-done-dir | /mr-history/tmp | Directory where history files are written by MapReduce jobs. |
| mapreduce.jobhistory.done-dir | /mr-history/done | Directory where history files are managed by the MR JobHistory Server. |
The HDFS and the YARN components are rack-aware.
The NameNode and the ResourceManager obtains the rack information of the slaves in the cluster by invoking an API resolve in an administrator configured module.
The API resolves the DNS name (also IP address) to a rack id.
The site-specific module to use can be configured using the configuration item topology.node.switch.mapping.impl. The default implementation of the same runs a script/command configured using topology.script.file.name. If topology.script.file.name is not set, the rack id /default-rack is returned for any passed IP address.
Hadoop provides a mechanism by which administrators can configure the NodeManager to run an administrator supplied script periodically to determine if a node is healthy or not.
Administrators can determine if the node is in a healthy state by performing any checks of their choice in the script. If the script detects the node to be in an unhealthy state, it must print a line to standard output beginning with the string ERROR. The NodeManager spawns the script periodically and checks its output. If the script's output contains the string ERROR, as described above, the node's status is reported as unhealthy and the node is black-listed by the ResourceManager. No further tasks will be assigned to this node. However, the NodeManager continues to run the script, so that if the node becomes healthy again, it will be removed from the blacklisted nodes on the ResourceManager automatically. The node's health along with the output of the script, if it is unhealthy, is available to the administrator in the ResourceManager web interface. The time since the node was healthy is also displayed on the web interface.
The following parameters can be used to control the node health monitoring script in conf/yarn-site.xml.
| Parameter | Value | Notes |
|---|---|---|
| yarn.nodemanager.health-checker.script.path | Node health script | Script to check for node's health status. |
| yarn.nodemanager.health-checker.script.opts | Node health script options | Options for script to check for node's health status. |
| yarn.nodemanager.health-checker.script.interval-ms | Node health script interval | Time interval for running health script. |
| yarn.nodemanager.health-checker.script.timeout-ms | Node health script timeout interval | Timeout for health script execution. |
The health checker script is not supposed to give ERROR if only some of the local disks become bad. NodeManager has the ability to periodically check the health of the local disks (specifically checks nodemanager-local-dirs and nodemanager-log-dirs) and after reaching the threshold of number of bad directories based on the value set for the config property yarn.nodemanager.disk-health-checker.min-healthy-disks, the whole node is marked unhealthy and this info is sent to resource manager also. The boot disk is either raided or a failure in the boot disk is identified by the health checker script.
Typically you choose one machine in the cluster to act as the NameNode and one machine as to act as the ResourceManager, exclusively. The rest of the machines act as both a DataNode and NodeManager and are referred to as slaves.
List all slave hostnames or IP addresses in your conf/slaves file, one per line.
Hadoop uses the Apache log4j via the Apache Commons Logging framework for logging. Edit the conf/log4j.properties file to customize the Hadoop daemons' logging configuration (log-formats and so on).
Once all the necessary configuration is complete, distribute the files to the HADOOP_CONF_DIR directory on all the machines.
To start a Hadoop cluster you will need to start both the HDFS and YARN cluster.
Format a new distributed filesystem:
$ $HADOOP_PREFIX/bin/hdfs namenode -format <cluster_name>
Start the HDFS with the following command, run on the designated NameNode:
$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs start namenode
Run a script to start DataNodes on all slaves:
$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs start datanode
Start the YARN with the following command, run on the designated ResourceManager:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR start resourcemanager
Run a script to start NodeManagers on all slaves:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR start nodemanager
Start a standalone WebAppProxy server. If multiple servers are used with load balancing it should be run on each of them:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh start proxyserver --config $HADOOP_CONF_DIR
Start the MapReduce JobHistory Server with the following command, run on the designated server:
$ $HADOOP_PREFIX/sbin/mr-jobhistory-daemon.sh start historyserver --config $HADOOP_CONF_DIR
Stop the NameNode with the following command, run on the designated NameNode:
$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs stop namenode
Run a script to stop DataNodes on all slaves:
$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs stop datanode
Stop the ResourceManager with the following command, run on the designated ResourceManager:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR stop resourcemanager
Run a script to stop NodeManagers on all slaves:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR stop nodemanager
Stop the WebAppProxy server. If multiple servers are used with load balancing it should be run on each of them:
$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh stop proxyserver --config $HADOOP_CONF_DIR
Stop the MapReduce JobHistory Server with the following command, run on the designated server:
$ $HADOOP_PREFIX/sbin/mr-jobhistory-daemon.sh stop historyserver --config $HADOOP_CONF_DIR
Once all the necessary configuration is complete, distribute the files to the HADOOP_CONF_DIR directory on all the machines.
This section also describes the various Unix users who should be starting the various components and uses the same Unix accounts and groups used previously:
To start a Hadoop cluster you will need to start both the HDFS and YARN cluster.
Format a new distributed filesystem as hdfs:
[hdfs]$ $HADOOP_PREFIX/bin/hdfs namenode -format <cluster_name>
Start the HDFS with the following command, run on the designated NameNode as hdfs:
[hdfs]$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs start namenode
Run a script to start DataNodes on all slaves as root with a special environment variable HADOOP_SECURE_DN_USER set to hdfs:
[root]$ HADOOP_SECURE_DN_USER=hdfs $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs start datanode
Start the YARN with the following command, run on the designated ResourceManager as yarn:
[yarn]$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR start resourcemanager
Run a script to start NodeManagers on all slaves as yarn:
[yarn]$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR start nodemanager
Start a standalone WebAppProxy server. Run on the WebAppProxy server as yarn. If multiple servers are used with load balancing it should be run on each of them:
[yarn]$ $HADOOP_YARN_HOME/bin/yarn start proxyserver --config $HADOOP_CONF_DIR
Start the MapReduce JobHistory Server with the following command, run on the designated server as mapred:
[mapred]$ $HADOOP_PREFIX/sbin/mr-jobhistory-daemon.sh start historyserver --config $HADOOP_CONF_DIR
Stop the NameNode with the following command, run on the designated NameNode as hdfs:
[hdfs]$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs stop namenode
Run a script to stop DataNodes on all slaves as root:
[root]$ $HADOOP_PREFIX/sbin/hadoop-daemon.sh --config $HADOOP_CONF_DIR --script hdfs stop datanode
Stop the ResourceManager with the following command, run on the designated ResourceManager as yarn:
[yarn]$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR stop resourcemanager
Run a script to stop NodeManagers on all slaves as yarn:
[yarn]$ $HADOOP_YARN_HOME/sbin/yarn-daemon.sh --config $HADOOP_CONF_DIR stop nodemanager
Stop the WebAppProxy server. Run on the WebAppProxy server as yarn. If multiple servers are used with load balancing it should be run on each of them:
[yarn]$ $HADOOP_YARN_HOME/bin/yarn stop proxyserver --config $HADOOP_CONF_DIR
Stop the MapReduce JobHistory Server with the following command, run on the designated server as mapred:
[mapred]$ $HADOOP_PREFIX/sbin/mr-jobhistory-daemon.sh stop historyserver --config $HADOOP_CONF_DIR
Once the Hadoop cluster is up and running check the web-ui of the components as described below:
| Daemon | Web Interface | Notes |
|---|---|---|
| NameNode | http://nn_host:port/ | Default HTTP port is 50070. |
| ResourceManager | http://rm_host:port/ | Default HTTP port is 8088. |
| MapReduce JobHistory Server | http://jhs_host:port/ | Default HTTP port is 19888. |