Hortonworks Hadoop System Admin Guide 20130819

docs.hortonworks.
com
Hortonworks Data Platform Aug 19, 2013
ii
Hortonworks Data Platform : System Administration Guides
Copyright 2012, 2013 Hortonworks, Inc. Some rights reserved.
The Hortonworks Data Platform, powered by Apache Hadoop, is a massively scalable and 100% open
source platform for storing, processing and analyzing large volumes of data. It is designed to deal with
data from many sources and formats in a very quick, easy and cost-effective manner. The Hortonworks
Data Platform consists of the essential set of Apache Hadoop projects including MapReduce, Hadoop
Distributed File System (HDFS), HCatalog, Pig, Hive, HBase, Zookeeper and Ambari. Hortonworks is the
major contributor of code and patches to many of these projects. These projects have been integrated and
tested as part of the Hortonworks Data Platform release process and installation and configuration tools
have also been included.
Unlike other providers of platforms built using Apache Hadoop, Hortonworks contributes 100% of our
code back to the Apache Software Foundation. The Hortonworks Data Platform is Apache-licensed and
completely open source. We sell only expert technical support, training and partner-enablement services.
All of our technology is, and will remain free and open source.
Please visit the Hortonworks Data Platform page for more information on Hortonworks technology. For
more information on Hortonworks services, please visit either the Support or Training page. Feel free to
Contact Us directly to discuss your specific needs.
Except where otherwise noted, this document is licensed under
Creative Commons Attribution ShareAlike 3.0 License.
http://creativecommons.org/licenses/by-sa/3.0/legalcode
Hortonworks Data Platform (HDP) and any of its components are not anticipated to be combined with any hardware, software or
data, except as expressly recommended in this documentation.
iii
Table of Contents
1. Configuring Rack Awareness on HDP .......................................................................... 1
1.1. Create a Rack Topology Script .......................................................................... 1
1.2. Add Properties to core-site.xml ......................................................................... 2
1.3. Restart HDFS and MapReduce .......................................................................... 2
1.4. Verify Rack Awareness ..................................................................................... 2
2. High Availability for Hadoop ....................................................................................... 3
3. High Availability for HDP Master Services Using VMware ............................................. 5
3.1. Use Cases and Fail Over Scenarios ..................................................................... 5
3.2. Supported Operating Systems ........................................................................... 6
3.3. Configuration For Physical Servers .................................................................... 6
3.4. Software Configuration .................................................................................... 6
3.4.1. Configure a vSphere HA cluster ............................................................. 7
3.4.2. Install HDP ............................................................................................. 8
3.4.3. Configure NameNode for automatic fail over ......................................... 9
3.4.4. Validate autostart configuration .......................................................... 10
3.4.5. Configure JobTracker for automatic fail over ........................................ 10
3.4.6. Validate JobTracker autostart configuration ......................................... 11
3.4.7. Enable vSphere for HA ......................................................................... 12
3.4.8. Validate NameNode High Availability ................................................... 13
3.5. Administration Guide for Highly Available NameNode ..................................... 18
3.5.1. NameNode shutdown for planned maintenance ................................... 18
3.5.2. Starting the NameNode ....................................................................... 18
3.5.3. Reconfiguring HDFS ............................................................................. 19
3.5.4. Tuning parameters for your environment ............................................. 19
3.6. References ...................................................................................................... 21
4. High Availability for HDP Master Services Using Red Hat ............................................ 22
4.1. Use Cases and Fail Over Scenarios ................................................................... 22
4.1.1. Supported use cases ............................................................................ 22
4.1.2. Supported fail over scenarios ............................................................... 23
4.2. Typical HDP HA Cluster ................................................................................... 23
4.3. Prerequisites ................................................................................................... 24
4.3.1. Hardware Prerequisites ........................................................................ 24
4.3.2. Software Prerequisites ......................................................................... 25
4.4. Install HDP Hadoop Core Packages ................................................................. 27
4.5. Deploy HDP HA Configurations ....................................................................... 27
4.6. Configure HA for NameNode Service .............................................................. 30
4.6.1. Install NameNode monitoring component ............................................ 30
4.6.2. Configure NameNode service in clustering configuration ....................... 30
4.7. Configure JobTracker HA for RHEL Cluster ...................................................... 31
4.7.1. Install JobTracker monitoring component ............................................. 31
4.7.2. Configure HDP JobTracker in clustering configuration ........................... 32
4.8. Distribute Cluster Configuration ...................................................................... 33
4.9. Validate Cluster Fail Over ................................................................................ 33
4.9.1. Validate NameNode restart on primary machine .................................. 34
4.9.2. Validate NameNode fail over during soft reboot .................................. 34
4.9.3. Validate NameNode fail over during hard reboot ................................. 34
5. High Availability for Hive Metastore .......................................................................... 35
5.1. Use Cases and Fail Over Scenarios ................................................................... 35
iv
5.2. Software Configuration .................................................................................. 36
5.2.1. Install HDP ........................................................................................... 36
5.2.2. Validate configuration ......................................................................... 36
6. Upgrade HDP Manually ............................................................................................. 37
6.1. Getting Ready to Upgrade .............................................................................. 37
6.2. Upgrade Hadoop ............................................................................................ 39
6.3. Upgrade ZooKeeper and HBase ...................................................................... 41
6.4. Upgrade Hive and HCatalog ........................................................................... 42
6.5. Upgrade Oozie ............................................................................................... 42
6.6. Upgrade WebHCat (Templeton) ..................................................................... 43
6.7. Upgrade Pig ................................................................................................... 44
6.8. Upgrade Sqoop .............................................................................................. 44
6.9. Upgrade Flume ............................................................................................... 45
6.9.1. Validate Flume ..................................................................................... 45
6.10. Upgrade Mahout .......................................................................................... 46
6.10.1. Mahout Validation ............................................................................. 46
7. Manually Add Slave Nodes to HDP Cluster ................................................................. 47
7.1. Prerequisites ................................................................................................... 47
7.2. Add DataNodes or TaskTrackers ..................................................................... 48
7.3. Add HBase RegionServer ................................................................................ 51
7.4. Optional - Configure Monitoring Using Ganglia ............................................... 53
7.5. Optional - Configure Cluster Alerting Using Nagios .......................................... 54
8. Decommission Slave Nodes ........................................................................................ 55
8.1. Prerequisites ................................................................................................... 55
8.2. Decommission DataNodes or TaskTrackers ...................................................... 55
8.2.1. Decommission DataNodes .................................................................... 56
8.2.2. Decommission TaskTrackers ................................................................. 56
8.3. Decommission HBase RegionServers ................................................................ 57
9. HDP Logfiles .............................................................................................................. 59
9.1. HDP Log Locations ......................................................................................... 59
9.2. HDP Log Format ............................................................................................. 60
9.3. HDP Backups .................................................................................................. 60
10. WebHDFS Administrator Guide ................................................................................ 61
v
List of Figures
2.1. FullStackHA.jpg ........................................................................................................ 4
vi
List of Tables
3.1. Custom properties for hdfs-site section ..................................................................... 8
3.2. Custom properties for mapred-site section ................................................................ 8
3.3. Custom properties for core-site section ..................................................................... 9
4.1. Parameters for configuring NameNode service in clustering configuration ................ 31
4.2. Parameters for configuring NameNode service in clustering configuration ................ 32
7.1. core-site.xml ........................................................................................................... 49
7.2. hdfs-site.xml ........................................................................................................... 49
7.3. mapred-site.xml ...................................................................................................... 49
7.4. taskcontroller.cfg .................................................................................................... 50
7.5. zoo.cfg ................................................................................................................... 52
7.6. hbase-site.xml ......................................................................................................... 52
9.1. Dependencies on HDP components ........................................................................ 59
1
1. Configuring Rack Awareness on HDP
Use the following instructions to configure rack awareness on a HDP cluster:
1. Create a Rack Topology Script
2. Add Properties to core-site.xml
3. Restart HDFS and MapReduce
4. Verify Rack Awareness
1.1.Create a Rack Topology Script
Topology scripts are used by Hadoop to determine the rack location of nodes. This
information is used by Hadoop to replicate block data to redundant racks.
1. Create a topology script and data file.
Sample Topology Script
File name: rack-topology.sh
HADOOP_CONF=/etc/hadoop/conf
while [ $# -gt 0 ] ; do
nodeArg=$1
exec< ${HADOOP_CONF}/topology.data
result=""
while read line ; do
ar=( $line )
if [ "${ar[0]}" = "$nodeArg" ] ; then
result="${ar[1]}"
fi
done
shift
if [ -z "$result" ] ; then
echo -n "/default/rack "
else
echo -n "$result "
fi
done
Sample Topology Data File
File name: topology.data
hadoopdata1.ec.com /dc1/rack1
hadoopdata1 /dc1/rack1
10.1.1.1 /dc1/rack2
2. Copy both of these files to /etc/hadoop/conf.
3. Run the rack-topology.sh script to ensure that it returns the correct rack
information for each host.
2
1.2.Add Properties to core-site.xml
1. Stop HDFS using the instructions on this page.
2. Add the following properties to core-site.xml:
<property>
<name>topology.script.file.name</name>
<value>/etc/hadoop/conf/rack-topology.sh</value>
</property>
<property>
<name>topology.script.number.args</name>
<value>1</value>
</property>
1.3.Restart HDFS and MapReduce
Restart HDFS and MapReduce using the instructions on this page.
1.4.Verify Rack Awareness
After the services have started, verify that rack awareness has been activated by looking in
the NameNode logs located in /var/log/hadoop/hdfs/ (for example: hadoop-hdfs-
namenode-sandbox.log). You should see an entry like this:
014-01-13 15:58:08,495 INFO org.apache.hadoop.net.NetworkTopology: Adding a
new node: /rack01/<ipaddress>
Also, the Hadoop fsck command should return something like the following (if there are
two racks):
Status: HEALTHY
Total size: 123456789 B
Total dirs: 0
Total files: 1
Total blocks (validated): 1 (avg. block size 123456789 B)
Minimally replicated blocks: 1 (100.0 %)
Over-replicated blocks: 0 (0.0 %)
Under-replicated blocks: 0 (0.0 %)
Mis-replicated blocks: 0 (0.0 %)
Default replication factor: 3
Average block replication: 3.0
Corrupt blocks: 0
Missing replicas: 0 (0.0 %)
Number of data-nodes: 40
Number of racks: 2
FSCK ended at Mon Jan 13 17:10:51 UTC 2014 in 1 milliseconds
3
2. High Availability for Hadoop
This section provides information on the various components of the Apache Hadoop
ecosystem and setting them up for high availability.
HDPs Full-Stack HA Architecture
Hortonworks Data Platform (HDP) is an open source distribution powered by Apache
Hadoop. HDP provides you with the actual Apache-released versions of the stack with
all the necessary bug fixes to make all the components in the stack interoperable in your
production environments. This stack uses multiple master services whose failure would
cause functionality outage in your Hadoop cluster. Hortonworks Full-Stack High Availability
architecture provides a common framework to make all the master services resilient to
failures.
HDP uses industry proven HA technologies in order to provide a reliable HA solution.
The Hadoop stack contains multiple services (HDFS, MapReduce, HBase, etc.) and each
of these services have their own co-dependencies. A client application, that interacts with
Hadoop, can depend on one or more of these services. A highly available Hadoop platform
must ensure that the NameNode master service as well as client applications are resilient to
critical failure services. Hortonworks Full-Stack HA architecture considers this global view.
Also see. Hortonworks blog on NameNode HA with Hadoop 1.0 The HDP HA architecture
has the following key properties:
It provides high availability for the NameNode master daemon service.
When the NameNode master daemon fails over, the HA solution initiates the following
actions:
Dependent services (like JobTracker) automatically detect the failure or fail over of the
co-dependent component (NameNode) and these dependent services pause, retry,
and recover the failed service. (For example, the JobTracker does not launch new jobs
or kill jobs that have been waiting for the NameNode.)
Applications running inside and outside the Hadoop cluster also automatically pause
and retry their connection to the failed service.
The above actions are highlighted in the following illustration. This illustration shows how
HDFS clients and MapReduce services (Jobtracker daemon) handle the NameNode fail over.
4
Figure2.1.FullStackHA.jpg
To configure High Availability for your Hadoop cluster see:
High Availability using Red Hat
High Availability using VMWare
5
3. High Availability for HDP Master
Services Using VMware
This document is intended for system administrators for configuring the Hadoop
NameNode (NN) and JobTracker (JT) services for High Availability and also for configuring
clients and other Hadoop services to be resilient while the NameNode or JobTracker fails
over.
The Hortonworks Data Platform (HDP) High Availability (HA) kit for VMware provides
enterprise grade high availability of the NameNode and JobTracker Service as part of the
HDP cluster.
To learn more, see HDP's Full-Stack HA Architecture.
In this document:
Use Cases and Fail Over Scenarios
Supported Operating Systems
Configuration For Physical Servers
Software Configuration
Enable vSphere for HA
Validate NameNode High Availability
Administration Guide for Highly Available NameNode
References
3.1.Use Cases and Fail Over Scenarios
This section provides information on the use cases and fail over scenarios for high
availability of the HDP Master Services (NameNode and JobTracker).
Use Cases
This solution enables the following Hadoop system administration use cases:
Planned downtime of the HDP Master Service (for maintenance tasks like software or
hardware upgrade)
Unplanned failure of the HDP Master Service
Fail over scenarios
The solution deals with the following faults:
HDP Master service failure
6
HDP Master JVM failure
Hung HDP Master daemon or hung operating system
HDP Master operating system failure
Virtual machine failure
ESXi host failure
Failure of the NIC cards on ESXi hosts.
Network failure between ESXi hosts.
Note
Some double faults are not handled (such as failure of multiple ESXi hosts).
3.2.Supported Operating Systems
The following operating systems for the Virtual Machines are supported:
Red Hat Enterprise Linux (RHEL) v5.7, v5.8, v6.2, v6.3, and later
CentOS v5.7, v5.8, v6.2, v6.3, and later
3.3.Configuration For Physical Servers
Ensure that your setup meets the following requirements:
The NameNode must run inside a virtual machine which is hosted on the vSphere HA
cluster.
The JobTracker must run inside a seperate virtual machine which is hosted on the
vSphere HA cluster.
The vSphere HA cluster must include a minimum of two ESXi server machines. See the
NameNode and JobTracker hardware recommendations available here.
The vSphere HA cluster must use shared storage.
Note
Shared storage stores the state of the NameNode edit journal and
checkpointed image. It provides ability to start the NameNode on another
physical machine in the vSphere HA Cluster. Additionally, vSphere uses the
shared state for the NameNodes root VM disk.
3.4.Software Configuration
Complete the following tasks to configure HDP Master Services (NameNode and/or
JobTracker) HA solution:
7
Configure a vSphere HA cluster
Install HDP
Configure HDP Master Service for automatic fail over
Validate autostart configuration
3.4.1.Configure a vSphere HA cluster
Complete the following configurations to configure your vSphere HA cluster:
Configure vSphere cluster [7]
Configure shared storage [7]
Configure vSphere cluster
To configure vSphere cluster, use the vSphere Management Console and complete the
following instructions. (Also see the VMware vSphere documentation.):
1. Create a vSphere HA cluster.
In the vCenter Client, select Inventory -> Hosts and Clusters.
In the left column, right-click the Datacenter and select New Cluster.
This step will create a vSphere HA cluster. (The vSphere HA cluster hosts the
NameNodes VM.)
2. Configure VM to host the NameNode
Within the vSphere HA cluster, create the virtual machines. Create one VM for each
HDP Master Service (One VM to host the NameNode process, and one VM to host the
JobTracker process).
Store the virtual machine disks (VMDK) on the data store located on the shared
storage device.
Specify the exact machine hardware configurations as required for configuring a
physical server for the NameNode and the JobTracker process.
Use the HDP Master service hardware recommendations available here.
Start these virtual machine and configure the network settings. You should now have
the IP address and FQDN for the virtual machines.
The virtual machines should be present on the same network where the rest of the
nodes of your Hadoop cluster are located.
Configure shared storage
Shared storage is required to host both the NameNode data directory and the VM
storage contents.
8
Use standard enterprise practices (such as configuring the shared storage to be RAIDed)
to ensure that your shared storage is reliable.
3.4.2.Install HDP
Use the following instructions to install HDP on your cluster hardware. Ensure that you
specify the virtual machines(configured in the previous section) as your NameNode and as
your JobTracker.
1. Download Hortonworks Data Platform (HDP) using the instructions provided here.
2. Use the fully qualified domain name (FQDN) of the virtual machine for configuring the
host names (see: Installing, Configuring, and Deploying the Cluster - Assign Masters.).
Note
HDP might not identify the NameNode or JobTracker VMs automatically and
it is therefore important to note down FQDN (IP address and DNS name) of
these VMs.
3. Specify shared storage for the NameNodes directories (see: HDP (Ambari) - Customize
Services).
4. Do not use the NameNode VM for running any other master daemon.
5. Do not use the JobTracker VM for running any other master daemon.
6. While deploying Ambari, add the following custom configuration parameters for Full-
Stack HA for both the NameNode and the JobTracker.
a. Add the following properties and values to custom hdfs-site section on the
dashboard:
Table3.1.Custom properties for hdfs-site section
Key Value Notes
dfs.client.retry.policy.enabled true Enable the HDFS client retry policy.
dfs.namenode.edits.toleration.length 8192 Configure protection for NameNode edit log.
dfs.safemode.extension 10 Configure safe mode extension time.
dfs.persist.blocks true Ensure that the allocated DFS blocks persist across
multiple fail overs.
dfs.blockreport.initialDelay 10 Configure delay for first block report.
b. Add the following properties and values to custom mapred-site section on the
dashboard:
Table3.2.Custom properties for mapred-site section
Key Value Notes
mapreduce.jt.hdfs.monitor.enable true Enable the JobTrackers safe mode functionality.
mapreduce.jobclient.retry.policy.enabled true Enable retry for JobTracker clients (when the
JobTracker is in safe mode).
9
Key Value Notes
mapred.jobtracker.restart.recovertrue Enable recovery of JobTrackers queue after it is
restarted.
c. Add the following properties and values to custom core-site section on the
dashboard:
Table3.3.Custom properties for core-site section
Key Value Notes
fs.checkpoint.period 3600 Configure checkpoint interval so that the checkpoint is
performed on an hourly basis.
3.4.3.Configure NameNode for automatic fail over
Complete the following instructions to configure automatic fail over:
Install and configure vSphere Monitoring Agent [9]
Configure the NameNode to start automatically on OS boot [10]
Install and configure vSphere Monitoring Agent
The NameNodes monitoring agent monitors the NameNode daemon and notifies the
vSphere if the NameNode daemon fails or becomes unstable. As soon as the vSphere
receives the notification, it triggers HA solution to restart NameNode VM on either the
same or a different ESXi host.
Follow the steps listed below to install and configure the monitoring agent:
1. On the NameNode VM, download the Hortonworks Data Platform High Availability Kit
for VMware from here and complete the following instructions:
Extract the archive file into a directory and navigate into the directory.
Install the NameNode monitoring functionality:
yum install hmonitor
yum install hmonitor-vsphere-monitoring-*
yum install hmonitor-vsphere-namenode-daemon
2. Configure the monitoring agent to point to the NameNode configurations.
Edit the /usr/lib/hadoop/monitor/vm-namenode.xml file and provide the
FQDN for your NameNode VM here:
<property>
<name>service.monitor.portprobe.host</name>
<value>$NameNode_FQDN</value>
<description> Port to probe via a telnet operation.</description>
</property>
Provide the HDFS port number.
10
<property>
<name>service.monitor.portprobe.port</name>
<value>$HDFS_filesystem_port</value>
</property>
Provide the FQDN and port information for the HDFS filesystem.
<property>
<name>service.monitor.webprobe.url</name>
<value>$http://NameNode_FQDN:NameNode_web_ui_port/</value>
<description> URL to get.</description>
</property>
Configure the NameNode to start automatically on OS boot
The monitoring agent requires that the NameNode process should start automatically once
the virtual machine is bootstrapped.
To configure this, use the following command:
yum install hadoop-namenode
3.4.4.Validate autostart configuration
Use the following instructions to validate that both the NameNode service and the
monitoring agent automatically get started on the reboot of the NameNode machine:
1. Reboot the NameNode machine.
2. Verify that the NameNode service is up and running.
On the NameNode machine, list all the Java processes by invoking jps and look for
NameNode process:
$JDK_HOME/bin/jps
If the service is up and running, you should see a message similar to the one shown in
the following example:
1628 NameNode
3. Verify that the NameNode monitoring agent is up and running.
On the NameNode machine, list all the Java processes by invoking jps and look for
NameNode process:
$JDK_HOME/bin/jps
1814 NameNodeMonitor
3.4.5.Configure JobTracker for automatic fail over
Complete the following instructions to configure automatic fail over:
11
Install and configure vSphere Monitoring Agent [11]
Configure the JobTracker to start automatically on OS boot [11]
Install and configure vSphere Monitoring Agent
The JobTrackers monitoring agent monitors the JobTracker daemon and notifies the
vSphere if the JobTracker daemon fails or becomes unstable. As soon as the vSphere
receives the notification, it triggers HA solution to restart JobTracker VM on either the
same or a different ESXi host.
Follow the steps listed below to install and configure the monitoring agent:
1. On the JobTracker VM, download the Hortonworks Data Platform High Availability Kit
for VMware from here and complete the following instructions:
Extract the archive file into a directory and navigate into the directory.
Install the JobTracker monitoring functionality:
yum install hmonitor
yum install hmonitor-vsphere-monitoring-*
yum install hmonitor-vsphere-jobtracker-daemon
2. Configure the monitoring agent to point to the JobTracker configurations.
Edit the /usr/lib/hadoop/monitor/vm-jobtracker.xml file and provide the
FQDN for your JobTracker VM here:
<property>
<name>service.monitor.portprobe.host</name>
<value>$JobTracker_FQDN</value>
</property>
Provide the JobTracker port number.
<property>
<name>service.monitor.portprobe.port</name>
<value>$JobTracker_port</value>
</property>
Configure the JobTracker to start automatically on OS boot
The monitoring agent requires that the JobTracker process should start automatically once
the virtual machine is bootstrapped.
To configure this, use the following command:
yum install hadoop-jobtracker
3.4.6.Validate JobTracker autostart configuration
Use the following instructions to validate that both the JobTracker service and the
monitoring agent automatically get started on the reboot of the JobTracker machine:
12
1. Reboot the JobTracker machine.
2. Verify that the JobTracker service is up and running.
On the JobTracker machine, list all the Java processes by invoking jps and look for
JobTracker process:
$JDK_HOME/bin/jps
1628 JobTracker
3. Verify that the JobTracker monitoring agent is up and running.
On the JobTracker machine, list all the Java processes by invoking jps and look for
JobTracker process:
$JDK_HOME/bin/jps
1814 JobTrackerMonitor
3.4.7.Enable vSphere for HA
Use the following instructions to modify the settings for the vSphere HA cluster created in
the previous section:
1. Right click on the cluster in vCenter and select Edit Settings.
2. In the Cluster Features panel, enable the Turn on vSphere HA option.
3. In the left navigation menu, select the vSphere HA option and configure the following
parameters.
Enable the Enable Host Monitoring option.
In the same panel, enable Admission Control.
13
Under Admission Control Policy section, select Host failures the cluster tolerates
and set the number of failures as 1.
4. In the left navigation menu, select the Virtual Machine Options and configure the
following parameters.
Set the VM restart priority to High.
Set the Host Isolation response to Leave powered on.
5. In the left navigation menu, select the VM Monitoring option and configure the
following parameters:
Set the VM Monitoring to VM and Application Monitoring.
Set the Monitoring sensitivity slider to High.
3.4.8.Validate NameNode High Availability
It is critical to verify that the NameNode HA infrastructure is functional.
To commission a NameNode HA cluster, we recommend performing explicit tests of
vSphere's ability to detect a failed NameNode and to react to that failure. These tests
should be performed in addition to any other commissioning tests.
To aid in this commissioning process, the HA monitor RPM includes the Hortonworks
Application Monitor (HAM) application. HAM can be used to monitor and restart a
remote NameNode. To validate NameNode High Availability, ensure that you follow the
instructions as outlined below:
Install and configure HAM
14
Invoke HAM application
Validate the failover behavior
3.4.8.1.Install and configure HAM
Typically, HAM can be run on the NameNode. However, we recommend that you also
install HAM on other nodes in the cluster because the NameNode will be restarted as part
of the validation tests. Use the following command to install HAM:
For RHEL v5.x:
rpm -ivh hmonitor-vsphere-namenode-daemon-*.rpm
For RHEL v6.x:
rpm -ivh hmonitor-vsphere-namenode-daemon-*.rpm
3.4.8.2.Invoke HAM application
Use the following command to invoke the HAM application:
/usr/lib/hadoop/monitor/ham.sh -conf hadoop-site.xml
This displays a GUI that lists the current values and state of the NameNode and Job Tracker.
You can use this GUI to trigger blocking and non-blocking HDFS operations.
Blocking operations will block until the NameNode has failed over. The blocking operations
use the dfs.retry properties. Non-blocking operations will fail if the NameNode is down,
but these operations start succeeding once the failover NameNode is live.
3.4.8.3.Validate the fail over behavior
Use the following tests to verify fail over behavior. (These tests can also be used to verify
that the availability monitoring can be suspended for administrative tasks.)
Verify that NameNode failure triggers the fail over [14]
Verify that a hung NameNode triggers the fail over [15]
Verify that ESXi server failure triggers the fail over [16]
Verify that no fail over is triggered on planned shutdown of the monitor service [16]
Verify that the monitor provides a bootstrap period before reporting that the
NameNode is not live [17]
Verify that no fail over is triggered when the NameNode enters the safe mode [17]
Verify that NameNode failure triggers the fail over
1. Start the NameNode VM and run the HAM application configured to work with this
NameNode.
2. In HAM, start blocking LS operations.
15
3. SSH to the NameNode VM and terminate the NameNode process.
service hadoop-namenode stop
Alternatively, identify the NameNode process (jps -v) and issue kill -9 command.
4. Ensure that you see the following expected results:
In HAM, the NameNode status area (at the top of the application) should display
offline status for NameNode. The main area should also stop displaying any new text
(this indicates that the file system operations are blocked).
In the vSphere Management UI, the vSphere should terminate the NameNode VM
within 60-90 seconds and must start a new instance.
Once the NameNode service restarts, its status must be displayed in both the vSphere
UI and in the status indicator of HAM.
The blocking operations started in HAM must now continue. The fail over should not
affect the client except for the pause during fail over.
SSH to the NameNode VM again and verify that the host name, IP address, and SSH
host key have not changed.
Verify that a hung NameNode triggers the fail over
This test verifies that the VM does not fail immediately after the NameNode process is
hung. The monitor considers this time period as a Garbage Collection-related pause. The
monitor provides a (configurable) period of grace time before it terminates the hung
NameNode process.
1. Start the NameNode VM and run the HAM application configured to work with this
NameNode.
2. In HAM, start non-blocking operations.
3. SSH to the NameNode VM and identify the NameNode process.
jps -v| grep namenode
4. Suspend the NameNode process.
kill -19 namenode-process-id-here
In HAM, the NameNode status area must indicate hung status. The non-blocking
operations, that are initiated, will now appear to be blocked (the hung NameNode
prevents these operations from completing).
In the vSphere Management UI, the vSphere should terminate the NameNode VM and
start a new instance within a delay of approximately 2-3 minutes.
In HAM, the NameNode status area should indicate offline status. The non-blocking
operations should now report failure.
16
The operations started in HAM will now start succeeding.
This test can be repeated when HAM performs blocking operations.
In this case, the active filesystem operation (the operation when the NameNode was
suspended) will fail when the NameNode is restarted and reported as such. This failure
happens because the open socket connection breaks and these connections are not
preserved during a fail over.
Verify that ESXi server failure triggers the fail over
This test verifies that the HA solution detects the failures of the physical hardware and also
trigger fail overs.
1. Start the NameNode VM in an ESXi server that is not running any other VMs.
2. Run the HAM application configured to work against this NameNode.
3. In the HAM, start blocking LS operations.
4. Initiate a power down of the ESXi server.
The main area should stop displaying new text - this indicates that the file system
operations are blocked.
In the vSphere management UI, once the loss of the ESXi server is detected, the
NameNode VM is re-instantiated on one of the remaining ESXi servers.
The blocked LS operation started in HAM should now continue without failures.
Verify that no fail over is triggered on planned shutdown of the monitor service
This test verifies that if the monitor service is shut down the fail over is not triggered. The
NameNode can now be manipulated as part of planned management operations.
1. Start the NameNode VM and SSH to the NameNode VM.
2. Terminate the monitor process.
service hmonitor-namenode-monitor stop
Note
A kill -9 command for the monitor is not a graceful shutdown and will trigger
fail over.
3. Terminate the NameNode process.
17
In the vSphere Management UI, the NameNode VM should be live.
The fail over should not be initiated by vSphere.
In HAM, the NameNode status area should indicate offline status. The non-blocking
operations should now report failure.
The SSH connection must not be broken and the VM should be live.
5. Restart the monitor process.
service hmonitor-namenode-monitor start
6. Restart the NameNode process.
service hadoop-namenode start
7. The NameNode health should be monitored and failures should trigger fail over.
Verify that the monitor provides a bootstrap period before reporting that the
NameNode is not live
This test verifies that the monitoring process includes a bootstrap period.
The bootstrap period ensures that the monitor will immediately not report a failure and
trigger a restart. Instead, the monitor provides the service a bootstrap period in which
probes are allowed to (initially fail). This bootstrap period is configurable (see: Tuning the
bootstrap timeout [19]).
2. Terminate the monitor process.
service hmonitor-namenode-monitor stop
4. Restart the monitor process.
service hmonitor-namenode-monitor start
Verify that no fail over is triggered when the NameNode enters the safe mode
This test verifies that the VM is not restarted if the NameNode enters safe mode. This
allows administration operations to be performed on a file system in safemode without
having to disable HA services.
2. Enter safe mode.
18
hadoop dfsadmin -safemode enter
In the vSphere UI, the NameNode VM should be live.
The SSH session should exist and the VM should be live.
5. The vSphere should identify the NameNode failure and should restart the VM.
Note
This test shows that even in safe mode the fail over is triggered if the
NameNode process is terminated. To avoid automatic restart for NameNode
after performing safe mode operations, use the service hmonitor-namenode-
monitor restart to restart the monitor service.
3.5.Administration Guide for Highly Available
NameNode
In this section:
NameNode shutdown for planned maintenance
Starting the NameNode
Reconfiguring HDFS
Tuning parameters for your environment
3.5.1.NameNode shutdown for planned maintenance
Use the following instructions to ensure controlled shutdown of the NameNode server:
1. Shutdown the NameNode monitoring agent.
service stop hmonitor-namenode-monitor
2. Shutdown the NameNode process.
service stop hadoop-namenode
3.5.2.Starting the NameNode
Use the following instructions to start the NameNode:
1. Start NameNode process.
service start hadoop-namenode
19
2. Start the NameNode monitoring agent and register it with vSphere.
service start hmonitor-namenode-monitor
3. Verify that the monitoring process has started.
service status hmonitor-namenode-monitor
3.5.3.Reconfiguring HDFS
Use the following instructions to reconfigure HDFS:
1. Disable HA on vSphere.
2. Shut down HDFS using the instructions available here.
3. Update the HDFS parameters using HDP. See the instructions here
Restart HDFS using HDP.See the instructions here
4. Enable HA on vSphere.
3.5.4.Tuning parameters for your environment
In this section:
Tuning parameters for your environment [19]
Tuning the reporting rate [20]
Tuning the probing rate [20]
Tuning for NameNode Garbage Collection [20]
Tuning parameters for your environment
When the VM starts, the HA Monitor waits for the NameNode to begin responding to file
system operations. During this "bootstrap phase, the HA monitor does not report startup
failures of NameNode probes to the HA infrastructure. The HA monitor exits the bootstrap
phase once all the probes succeed (from that point, the failure of a probe is reported as a
service failure).
The time limit of the bootstrap phase can be configured using the
service.monitor.bootstrap.timeout property:
<property>
<name>service.monitor.bootstrap.timeout</name>
<value>120000</value>
<description>
The time in milliseconds for the monitor to wait for the service to bootstrap
and
become available before it reports a failure to the management infrastructure
</description>
</property>
20
The timeout must be sufficiently long so that the monitored service is able to open its
network ports for external interaction. For the NameNode, the web page and IPC port
must be open.
The bootstrap time also needs to include the time required for the HDFS journal replay
operations. The bootstrap timeout value should be kept high if the filesystem is large and if
the secondary NameNode checkpointing time intervals are longer.
Tuning the reporting rate
The internal VM Monitor daemon sends heartbeat messages to vSphere to indicate that
the VM is alive. Use the following property, to modify the rate at which these heartbeats
are sent.
<property>
<name>service.monitor.report.interval</name>
<value>7000</value>
<description>
Interval in milliseconds for sending heartbeats to vSphere.
</description>
</property>
It is essential that the live VM sends a heartbeats to vSphere at least every thirty seconds.
A smaller reporting interval reduces the risk of missed heartbeats in case an Operating
System or Java related process hangs. However, a smaller reporting interval can also have
adverse effects - especially if the VM is overloaded. It is therefore strongly recommended
to address the root cause of the VM overload. If your VM is overloaded and becomes
unresponsive, we recommend that you either add more CPUs and RAM or rebalance VMs
across the cluster.
Tuning the probing rate
The Monitor daemon probes the health of the NameNode at a regular interval, and stops
sending heartbeats to vSphere when any of the probes fail. Use the following property to
change the rate of probes:
<property>
<name>service.monitor.probe.interval</name>
<description>
Time in milliseconds between the last probe cycle ending and the new one
beginning.
The shorter this cycle, the faster failures are detected, but more CPU,
network,
and server load can be generated.
</description>
</property>
The smaller the interval between probes, the faster it becomes to detect and report service
failures. Although, this might increase the load on the service and the CPU slightly, but even
with a very short probing interval, vSphere will not trigger VM restart for at least thirty
seconds after the probe failure.
Tuning for NameNode Garbage Collection
21
The NameNode process can appear hung during Garbage Collection event. To prevent this
from triggering immediate failover, a grace period is provided to the NameNode to resume
its operation. You can configure this grace period using the following property:
<property>
<name>service.monitor.probe.timeout</name>
<description>
Duration in milliseconds for the probe loop to be blocked, before it is
considered a liveness failure
</description>
</property>
A smaller value will cause the VM (where the hung NameNode process is running) faster,
but it increases the risk of incorrectly identifying a long GC-related pause as a hung process.
On larger clusters (with longer GC pauses), you can increase the value of this property.
3.6.References
High Availability for Hadoop using VMWare
22
4. High Availability for HDP Master
Services Using Red Hat
This section describes using Red Hat configurations to support high availability.
This document is intended for system administrators for configuring the Hadoop
NameNode (NN) and JobTracker (JT) service for High Availability and also for configuring
clients and other Hadoop services to be resilient while the NameNode or JobTracker fails
over.
The Hortonworks Data Platform (HDP) High Availability (HA) kit for Red Hat Enterprise
Linux (RHEL) provides enterprise grade high availability of the NameNode Service as part of
the HDP cluster.
In this document:
Use Cases and Fail Over Scenarios
Typical HDP HA Cluster
Prerequisites
Install HDP Hadoop Core Packages
Deploy HDP HA Configurations
Configure NameNode HA for RHEL Cluster
Configure JobTracker HA for RHEL Cluster
Distribute Cluster Configuration
Validate Cluster Fail Over
This section provides information on the following:
Supported use cases
Supported fail over scenarios
4.1.1.Supported use cases
This solution enables the following Hadoop system administration use cases:
Planned failure of the HDP Master Service (maintenance tasks like software or hardware
upgrade)
23
Unplanned failure of the HDP Master Service (hardware or OS failure, software bugs,
JVM related issues, etc.)
4.1.2.Supported fail over scenarios
The following failover scenarios are supported:
HDP Master Service service failure
HDP Master Service JVM failure
Hung HDP Master Service daemon or hung operating system
HDP Master Service operating system failure
HDP Master Service machine power failure
Failure of NIC cards on the HDP Master Service machine
Network failure for the HDP Master Service machine
4.2.Typical HDP HA Cluster
When you configure HDP HA solution using RHEL, your Hadoop cluster comprises of the
following two components:
RHEL HA cluster
Overall Hadoop cluster
Important
The RHEL HA cluster is a subset of the overall Hadoop cluster.
Typically, the overall Hadoop cluster must include the following types of machines:
The RHEL HA cluster machines. These machines must host only those master services that
require HA (in this case, the NameNode and the JobTracker).
Master machines that run other master services such as Hive Server 2, HBase master, etc.
Slave machines that run slave daemons like Datanodes, TaskTracker, RegionServers etc.
Client machines.
A network that connects all the machines.
The following illustrates the topology of a sample Hadoop cluster when configured for HDP
HA using RHEL:
24
4.3.Prerequisites
Complete the following prerequisites before configuring HA for your cluster:
1. Hardware prerequisites
2. Software prerequisites
4.3.1.Hardware Prerequisites
Ensure that you complete the following hardware prerequisites:
Shared Storage.
Shared storage is required for storing the NameNode metadata. Use a highly available
shared storage NFS device.
Ensure that you use a Power fencing device.
Note
Red Hat HA cluster utilizes power fencing to deal with network split-brain
events. Fencing guarantees the integrity of NameNode metadata. For more
information, see: Fencing Topology.
IP fail over with a floating IP
Ensure that additional static IPs are available for the HDP Master Services or the
cluster.
The IP must be a static reserved entry in your network DNS table. This IP will act as the
public IP for the HDP Master Service.
Note
Red Hat HA clustering utilizes a floating IP address for the NameNode and/
or the JobTracker service across the HA cluster. More details on using a
floating IP for RHEL are available here.
25
Hardware requirement for RHEL HA cluster
The RHEL HA cluster must have a minimum of two nodes.
The number of nodes in your HA cluster depends on the number of concurrent
node failures you want the HDP platform to withstand. The RHEL HA cluster can be
configured to include a maximum of 16 nodes. Choose hardware specs for the RHEL
HA Cluster nodes according to the NameNode hardware recommendations available
here.
4.3.2.Software Prerequisites
Ensure that you complete the following software prerequisites:
Configure RHEL HA cluster
Validate configurations for RHEL HA cluster
4.3.2.1.Configure RHEL HA Cluster
Use the following instructions:
1. Complete the prerequisites for High Availability Add-On package for RHEL. Use the
instructions available here (RHEL v5.x, RHEL v6.x).
2. Install the HA Add-On package for RHEL. Use the instructions available here (RHEL
v5.x,RHEL v6.x ).
Important
You can use the graphical user interface (GUI) to configure a RHEL v6.x
cluster configuration until you specify a Hadoop service configuration
(Deploy HDP HA Configurations). You must use the cluster.conf file
to specify the Hadoop service configuration. Once the Hadoop service
configuration is put in place, any changes made via the GUI will break the
configuration. You can still use the GUI to manage the HA NameNode
service -- start, stop, and move the service across cluster machines.
3. Ensure that the following cluster configurations are available on all the machines in your
RHEL HA cluster:
Cluster domain that specifies all the nodes in the RHEL HA cluster. See instructions
here (RHEL v5.x, RHEL v6.x).
Fail over domain. See instructions here (RHEL v5.x, RHEL v6.x).
Power Fencing device. See instructions here (RHEL v5.x, RHEL v6.x).
Add cluster service and resources (Floating IP and NFS mount). Ensure that you add
the <service domain> configurations and to add resources to the service group:
See instructions here (RHEL v5.x, RHEL v6.x).
When the above are configured, you will have a cluster.conf file similar to
the following sample configuration. (Note that this sample configuration does not
26
declare a true fencing device because that is specific to the environment. Modify the
configuration values to match your infrastructure environment.)
<?xml version="1.0"?>
<cluster config_version="8" name="rhel6ha">
<clusternodes>
<clusternode name="rhel6ha01" nodeid="1">
<fence>
<method name="1">
<device name="BinTrue"/>
</method>
</fence>
</clusternode>
<clusternode name="rhel6ha02" nodeid="2">
<fence>
<method name="1">
<device name="BinTrue"/>
</method>
</fence>
</clusternode>
</clusternodes>
<cman expected_votes="1" two_node="1"/>
<fencedevices>
<fencedevice agent="fence_bin_true" name="BinTrue"/>
</fencedevices>
<rm log_level="7">
<failoverdomains>
<failoverdomain name="HDPMaster" ordered="1" restricted="1">
<failoverdomainnode name="rhel6ha01" priority="1"/>
<failoverdomainnode name="rhel6ha02" priority="2"/>
</failoverdomain>
</failoverdomains>
<service domain="HDPMaster" name="NameNodeService"
recovery="relocate">
<ip address="10.10.10.89" sleeptime="10"/>
<netfs export="/hdp/nfs" force_unmount="1" fstype="nfs" host="10.10.
10.88" mountpoint="/hdp/hadoop/hdfs/nn" name="HDFS data" options="rw,soft,
nolock"/>
</service>
</rm>
</cluster>
4.3.2.2.Validate Configurations for RHEL HA Cluster
Use the following instructions to validate the configurations for RHEL HA cluster:
1. Validate that the floating IP address is available on the primary machine. (Primary
machine is the machine where the NameNode process is currently running).
ip addr show eth1
If the IP address is available, you should see a message (as shown in
the following example). In this example, the IP address is configured at
rheln1.hortonworks.local:
root@rheln1 ~]# ip addr show eth1
27
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state
UP qlen 1000
link/ether 00:0c:29:cb:ca:76 brd ff:ff:ff:ff:ff:ff
inet 172.16.204.10/24 brd 172.16.204.255 scope global eth1
inet 172.16.204.12/24 scope global secondary eth1
inet6 fe80::20c:29ff:fecb:ca76/64 scope link
valid_lft forever preferred_lft forever
2. Validate that the failover service (in the above example, the NameNodeService starts
on the secondary machine).
ip addr show eth3
3. Validate fail over for the IP address.
Shut down alternate host machines.
Ensure that the IP address fails over properly.
4.4.Install HDP Hadoop Core Packages
Use the following instructions:
1. Install Hadoop Core packages on the entire cluster using the instructions provided here.
2. Ensure that you complete the following set-up requirements for NameNode machines:
Use the floating IP for the NameNode network location ($DFS_NAME_DIR). Add a
DNS entry with a hostname that points to the floating IP address. Use this hostname
to specify the NameNode host address.
Configure the NameNode data directory as a child of the mounted shared storage
directory. For example, if you have the NFS directory mounted at /hdp/hdfs/nn,
then the NameNode data directory should be configured to be at the following
location: /hdp/hdfs/nn/data.
Replicate the above set-up of the NameNode machine on all the machines in your
RHEL HA cluster. Ensure that each machine in the RHEL HA cluster has identical HDP
packages and add-on libraries.
3. Ensure that you complete the following set-up requirements for JobTracker machines:
Use the floating IP for the JobTracker network location. Add a DNS entry with a
hostname that points to the floating IP address. Use this hostname to specify the
JobTracker host address.
Replicate the above set-up of the JobTracker machine on all the machines in your
RHEL HA cluster that you want to run the JobTracker on. Ensure that each machine in
the RHEL HA cluster has identical HDP packages and add-on libraries.
4.5.Deploy HDP HA Configurations
Use the instructions provided in this section to configure Full-Stack HA fail over resiliency for
the HDP clients.
28
Note
Your Hadoop configuration directories are defined during the HDP installation.
For details, see: Setting Up Hadoop Configuration.
1. Edit the $HADOOP_CONF_DIR/hdfs-site.xml file to add the following properties:
Enable the HDFS client retry policy.
<property>
<name>dfs.client.retry.policy.enabled</name>
<value>true</value>
<description> Enables HDFS client retry in case of NameNode failure.</
description>
</property>

Configure protection for NameNode edit log.
<property>
<name>dfs.namenode.edits.toleration.length</name>
<value>8192</value>
<description> Prevents corruption of NameNode edit log.</description>
</property>

Configure safe mode extension time.
<property>
<name>dfs.safemode.extension</name>
<value>10</value>
<description> The default value (30 seconds) is applicable for very large
clusters. For small to large clusters (upto 200 nodes), recommended value
is 10 seconds.</description>
</property>

Ensure that the allocated DFS blocks persist across multiple fail overs.
<property>
<name>dfs.persist.blocks</name>
<value>true</value>
<description>Ensure that the allocated DFS blocks persist across multiple
fail overs.</description>
</property>

Configure delay for first block report.
29
<property>
<name>dfs.blockreport.initialDelay</name>
<value>10</value>
<description> Delay (in seconds) for first block report.</description>
</property>

where $HADOOP_CONF_DIR is the directory for storing the Hadoop configuration files.
For example, /etc/hadoop/conf.
2. Modify the following property in the $HADOOP_CONF_DIR/core-site.xml file:
<property>
<name>fs.checkpoint.period</name>
<value>3600</value>
<description> The number of seconds between two periodic checkpoints.</
description>
</property>

This will ensure that the checkpoint is performed on an hourly basis.
3. Edit the $HADOOP_CONF_DIR/mapred-site.xml file to add the following properties:
Enable the JobTrackers safe mode functionality.
<property>
<name>mapreduce.jt.hdfs.monitor.enable</name>
<value>true</value>
<description> Enable the JobTracker to go into safe mode when the
NameNode is not responding.</description>
</property>

Enable retry for JobTracker clients (when the JobTracker is in safe mode).
<property>
<name>mapreduce.jobclient.retry.policy.enabled</name>
<value>true</value>
<description> Enable the MapReduce job client to retry job submission
when the JobTracker is in safe mode.</description>
</property>

Enable recovery of JobTrackers queue after it is restarted.
<property>
<name>mapred.jobtracker.restart.recover</name>
<value>true</value>
<description> Enable the JobTracker to recover its queue after it is
restarted.</description>
</property>

30
4.6.Configure HA for NameNode Service
Follow the tasks listed below to configure NameNode HA:
Install NameNode monitoring component
Configure NameNode service in clustering configuration
4.6.1.Install NameNode monitoring component
1. Stop all the HDP services that are currently running on your cluster using the instructions
provided here.
2. Stop the RHEL cluster NameNodeService service using the RHEL Cluster administration
tools (RHEL v5.x., RHEL v6.x.).
3. Install the NameNode monitoring component on all the nodes in your RHEL HA cluster.
Ensure that you have set up the HDP repository on the RHEL HA cluster nodes as part
of the HDP installation using the instructions provided here.
Install the RPMs.
For RHEL/CentOS 5
yum install hmonitor*.rpm
yum install hmonitor-resource-agent*.rpm

For RHEL/CentOS 6

4.6.2.Configure NameNode service in clustering
configuration
Edit the /etc/cluster/cluster.conf file to add the service domain specifications.
You can use the following sample configuration. (Note that this sample configuration is for
a small cluster and the timeouts for booting, probing, and stopping have been reduced to a
minimum.)
<service domain="HANameNode" name="NameNodeService" recovery="restart">
<netfs export="/hdp/hadoop-nfs" force_unmount="1" fstype="nfs"
host="10.10.10.88" mountpoint="/hdp/hadoop/hdfs/nn"
name="HDFS data" options="rw,soft,nolock"/>
<hadoop
__independent_subtree="1" __max_restarts="10"
31
__restart_expire_time="600" name="NameNode Process"
daemon="namenode" boottime="10000" probetime="10000"
stoptime="10000"
url="http://10.0.0.30:50070/dfshealth.jsp"
pid="/var/run/hadoop/hdfs/hadoop-hdfs-namenode.pid"
path="/"/>
</service>

The following table explains the parameters used in the above configuration:
Table4.1.Parameters for configuring NameNode service in clustering
configuration
Name Description Mandatory/Optional
daemon Name of the hadoop service which will be started by bin/
hadoop-daemon.
Mandatory
url URL to check for the service web page. This should be on the
floating IP address.
Mandatory
pid Process ID of the NameNode process. (Default: ) Optional
path Path under DFS to probe. (Default: /) Optional
boottime Time (in milliseconds) to allow for the service to boot up. This
must include any activities that take place before the service
web pages and IPC services are reachable. For the NameNode, it
must include the time for the edit log to be replayed. (Default:
180000)
Optional
probetime Time (in milliseconds) to allow for a process to respond to
liveness probes. This duration must be longer than the maximum
expected GC pause. (Default: 120000)
Optional
stoptime Time (in milliseconds) to allow for a clean shutdown before
forcibly killing a process. (Default: 60000)
Optional
4.7.Configure JobTracker HA for RHEL Cluster
Follow the tasks listed below to configure JobTracker HA:
Install JobTracker monitoring component
Configure JobTracer service in clustering configuration
4.7.1.Install JobTracker monitoring component
1. Stop all the HDP services that are currently running on your cluster using the instructions
provided here.
2. Stop the RHEL cluster JobTrackerService service using the RHEL Cluster administration
tools (RHEL v5.x., RHEL v6.x.).
3. Install the JobTracker monitoring component on all the nodes in your RHEL HA cluster.
Ensure that you have set up the HDP repository on the RHEL HA cluster nodes as part
of the HDP installation.
Use the instructions provided here.
32
Install the RPMs.

4.7.2.Configure HDP JobTracker in clustering configuration
Edit the /etc/cluster/cluster.conf file to add the service domain specifications.
You can use the following sample configuration. (Note that this sample configuration is for
a small cluster and the timeouts for booting, probing, and stopping have been reduced to a
minimum.)
<service domain="HAJobTracker" name="JobTrackerService" recovery="restart">
<hadoop
__restart_expire_time="600"
name="JobTracker Process"
daemon="jobtracker"
boottime="60000"
probetime="20000"
stoptime="30000"
url="http://10.0.0.30:50030/"
waitfs="true"
/>
<hadoop
__restart_expire_time="600"
name="HistoryServer Process"
daemon="historyserver"
boottime="60000"
probetime="20000"
stoptime="30000"
url="http://10.0.0.30:51111/"
waitfs="true"
/>

The following table explains the parameters used in the above configuration:
Table4.2.Parameters for configuring NameNode service in clustering
configuration
daemon Name of the hadoop service which will be started by bin/
hadoop-daemon.
Mandatory
url URL to check for the service web page. This should be on the
floating IP address.
Mandatory
pid Process ID of the Master Service process. (Default: ) Optional
boottime Time (in milliseconds) to allow for the service to boot up. This
must include any activities that take place before the service
web pages and IPC services are reachable. (Default: 180000)
Optional
33
probetime Time (in milliseconds) to allow for a process to respond to
liveness probes. This duration must be longer than the maximum
expected GC (Garbage Collection) pause. (Default: 120000)
Optional
stoptime Time (in milliseconds) to allow for a clean shutdown before
forcibly killing a process. (Default: 60000)
Optional
4.8.Distribute Cluster Configuration
1. Ensure that ricci is running on the current node and on all other nodes of the cluster.
service ricci status

This command should return the status message similar to the one shown in the
following example:
ricci (pid 9524) is running...

2. Distribute the configuration to other nodes in the cluster.
Execute the following command from the node that contains the cluster.conf file.
For RHEL v5.x:
ccs_tool update /etc/cluster/cluster.conf

For RHEL v6.x:
ccs -i -h host --sync --activate

3. Validate the updated configuration files.
SSH into another node in the cluster and execute the following command:
cat /etc/cluster/cluster.conf

This command should display the updated contents for the cluster.conf file.
4.9.Validate Cluster Fail Over
Use the following tests to validate successful fail over for the NameNode master daemon
service:
Validate NameNode restart on primary machine
Validate NameNode fail over during soft reboot
34
Validate NameNode fail over during hard reboot
4.9.1.Validate NameNode restart on primary machine
This test simulates the event where the NameNode master daemon dies on the primary
(current) machine. A successful fail over should restart the NameNode process on the
primary machine.
From the primary machine, perform the following command to shutdown the
namenode:
su - hdfs
/usr/lib/hadoop/bin/hadoop-daemon.sh --config $HADOOP_CONF_DIR stop namenode

Once the NameNode is shutdown, the monitor must detect the unavailability of the
NameNode and should attempt to restart it. This process will take approximately 30
seconds.
4.9.2.Validate NameNode fail over during soft reboot
This test simulates the event where the NameNode server becomes unavailable due to a
planned reboot of the primary NameNode machine. A successful fail over should be able to
relocate both the floating IP and the NameNode process to the secondary machine in the
cluster. When the primary machine completes the restart, the NameNode process should
again be relocated to the primary machine.
Perform a soft reboot of the server.
During the shutdown process, the other machine in the cluster will be notified and will
begin the migration of the service.
The service should be available on the secondary machine in the cluster until the primary
machine has completed its reboot.
4.9.3.Validate NameNode fail over during hard reboot
This test simulates the event when the NameNode server becomes unavailable due to an
unplanned reboot of the primary NameNode machine. A successful fail over should be
able to relocate both the floating IP and the NameNode process to the secondary machine
in the cluster. When the primary machine completes the restart, the NameNode process
should again be relocated to the primary machine.
Perform a hard shutdown of the primary NameNode machine.
Once the secondary machine in the cluster detects the outage, it must start the
NameNode process on the secondary machine.
The NameNode service should be available on the secondary machine until the primary
machine completes its startup routine.
35
5. High Availability for Hive Metastore
This document is intended for system administrators who need to configure the Hive
Metastore service for High Availability.
This section provides information on the use cases and fail over scenarios for high
availability (HA) in the Hive metastore.
Use Cases
The metastore HA solution is designed to handle metastore service failures. Whenever
a deployed metastore service goes down, metastore service can remain unavailable for
a considerable time until service is brought back up. To avoid such outages, deploy the
metastore service in HA mode.
Deployment Scenarios
We recommend deploying the metastore service on multiple boxes concurrently. Each Hive
metastore client will read the configuration property hive.metastore.uris to get a list
of metastore servers with which it can try to communicate.
<property>
<name> hive.metastore.uris </name>
<value> thrift://$Hive_Metastore_Server_Host_Machine_FQDN </value>
<description> A comma separated list of metastore uris on which metastore
service is running </description>
</property>

These metastore servers store their state in a MySQL HA cluster, which should be set up as
recommended in the whitepaper "MySQL Replication for Failover Protection."
In the case of a secure cluster, each of the metastore servers will additionally need to have
the following configuration property in its hive-site.xml file.
<property>
<name> hive.cluster.delegation.token.store.class </name>
<value> org.apache.hadoop.hive.thrift.DBTokenStore </value>
</property>

Fail Over Scenario
A Hive metastore client always uses the first URI to connect with the metastore server. In
case the metastore server becomes unreachable, the client will randomly pick up a URI from
the list and try connecting with that.
36
5.2.Software Configuration
Complete the following tasks to configure Hive HA solution:
Install HDP
Validate configuration
5.2.1.Install HDP
Use the following instructions to install HDP on your cluster hardware. Ensure that you
specify the virtual machine (configured in the previous section) as your NameNode.
1. Download Apache Ambari using the instructions provided here.
Note
Do not start the Ambari server until you have configured the relevant
templates as outlined in the following steps.
2. Edit the <master-install-machine-for-Hive-Metastore>/TODO file to add
the following properties:
a. Provide the URI for client to contact Metastore server. The following property can
have a comma separated list when your cluster has multiple Hive Metastore servers.
<property>
<name> hive.metastore.uris </name>
<value> thrift://$Hive_Metastore_Server_Host_Machine_FQDN </value>
<description> URI for client to contact metastore server </description>
</property>
b. Configure Hive cluster delegation token storage class.
<property>
<name> hive.cluster.delegation.token.store.class </name>
<value> org.apache.hadoop.hive.thrift.DBTokenStore </value>
</property>
3. Complete HDP installation.
Continue the Ambari installation process using the instructions provided here.
Complete the Ambari installation. Ensure that the installation was successful.
5.2.2.Validate configuration
Test various fail over scenarios to validate your configuration.
37
6. Upgrade HDP Manually
This document provides instructions on how to upgrade to the latest release of HDP from
an earlier HDP release. Use the following instructions to upgrade to the latest release of
HDP:
1. Getting Ready to Upgrade
2. Upgrade Hadoop
3. Upgrade ZooKeeper and HBase
4. Upgrade Hive and HCatalog
5. Upgrade Oozie
6. Upgrade WebHCat (Templeton)
7. Upgrade Pig
8. Upgrade Sqoop
9. Upgrade Flume
10.Upgrade Mahout
6.1.Getting Ready to Upgrade
1. Stop all services (including MapReduce) and client applications deployed on HDFS using
the instructions provided here.
2. Run the fsck command as instructed below and fix any errors. (The resulting file will
contain a complete block map of the file system.)
su $HDFS_USER
hadoop fsck / -files -blocks -locations > dfs-old-fsck-1.log
where $HDFS_USER is the HDFS Service user. For example, hdfs.
3. Use the following instructions to compare the status before and after the upgrade:
Note
The following commands must be executed by the user running the HDFS
service (by default, the user is hdfs).
a. Capture the complete namespace of the file system. Run recursive listing of the root
file system: )
su $HDFS_USER
hadoop dfs -lsr / > dfs-old-lsr-1.log
38
b. Run report command to create a list of DataNodes in the cluster.
su $HDFS_USER
hadoop dfsadmin -report > dfs-old-report-1.log
c. Copy all or unrecoverable data stored in HDFS to a local file system or to a backup
instance of HDFS.
d. Optionally, repeat the steps 3 (a) through 3 (c) and compare the results with the
previous run to verify that the state of the file system remains unchanged.
4. As an HDFS user, execute the following command to save namespace:
su $HDFS_USER
hadoop dfsadmin -safemode enter
hadoop dfsadmin -saveNamespace
5. Copy the following checkpoint files into a backup directory:
dfs.name.dir/edits
dfs.name.dir/image/fsimage
6. Stop the HDFS service using the instructions provided here. Verify that all the HDP
services in the cluster are stopped.
7. If you are upgrading Hive, back up the Hive database.
8. Configure the local repositories.
The standard HDP install fetches the software from a remote yum repository over the
Internet. To use this option, you must set up access to the remote repository and have
an available Internet connection for each of your hosts.
Note
If your cluster does not have access to the Internet, or you are creating a
large cluster and you want to conserve bandwidth, you can instead provide
a local copy of the HDP repository that your hosts can access. For more
information, see Deployment Strategies for Data Centers with Firewalls, a
separate document in this set.
The file you download is named hdp.repo. To function properly in the
system, it must be named HDP.repo. Once you have completed the mv of
the new repo file to the repos.d folder, make sure there is no file named
hdp.repo anywhere in your repos.d folder.
a. Upgrade the HDP repository on all hosts and replace the old repo file with the new
file.
39
From a terminal window, type:
For RHEL and CentOS 5
wget http://dev.hortonworks.com.s3.amazonaws.com/HDP/centos5/1.x/GA/
hdp.repo -O /etc/yum.repos.d/hdp.repo
For RHEL and CentOS 6
wget http://dev.hortonworks.com.s3.amazonaws.com/HDP/centos6/1.x/GA/
hdp.repo -O /etc/yum.repos.d/hdp.repo
For SLES 11
wget http://dev.hortonworks.com.s3.amazonaws.com/HDP/suse11/1.x/GA/
hdp.repo -O /etc/zypp/repos.d/hdp.repo
b. Confirm that the HDP repository is configured by checking the repo list.
For RHEL/CentOS:
yum repolist
For SLES:
zypper repos
6.2.Upgrade Hadoop
1. On all nodes:
For RHEL/CentOS:
yum upgrade hadoop*
For SLES:
zypper update hadoop*
2. Start HDFS.
a. Start NameNode. On the NameNode host machine, execute the following command:
sudo su -l $HDFS_USER -c "/usr/lib/hadoop/bin/hadoop-daemon.sh start
namenode -upgrade"

b. Start Secondary NameNode. On the Secondary NameNode host machine, execute the
following command:
secondarynamenode"

c. Start DataNodes. On all the DataNodes, execute the following command:
datanode"
40

d. Execute the following command on the NameNode machine:
hadoop dfsadmin -safemode wait
e. Track the status of the upgrade:
hadoop dfsadmin -upgradeProgress status
Continue tracking until you see:
Upgrade for version -44 has been completed.
Upgrade is not finalized.
Note
You finalize the upgrade later.
3. The upgrade is now fully functional but not yet finalized. Using the finalize comand
removes the previous version of the NameNode and DataNode's storage directories.
Important
Once the upgrade is finalized, the system cannot be rolled back. Usually
this step is not taken until a thorough testing of the upgrade has been
performed.
The upgrade must be finalized, however, before another upgrade can be performed.
To finalize the upgrade:
su $HDFS_USER
hadoop dfsadmin -finalizeUpgrade
4. Start MapReduce.
a. Start JobTracker. On the JobTracker host machine, execute the following command:
sudo su -l $MAPRED_USER -c "/usr/lib/hadoop/bin/hadoop-daemon.sh start
jobtracker"

b. Start JobHistory Server. On the JobHistory Server host machine, execute the following
command:
sudo su -l $MAPRED_USER -c "/usr/lib/hadoop/bin/hadoop-daemon.sh start
historyserver"

c. Start all TaskTrackers. On all the TaskTrackers, execute the following command:
sudo su -l $MAPRED_USER -c \"/usr/lib/hadoop/bin/hadoop-daemon.sh start
tasktracker\"
41

where $MAPRED_USER is the MapReduce Service user. For example, mapred.
6.3.Upgrade ZooKeeper and HBase
1. Install ZooKeeper. Execute the following command on all the ZooKeeper nodes:
For RHEL/CentOS:
yum downgrade zookeeper-3.4.5.1.3.0.0

For SLES:
zypper remove zookeeper
zypper update zookeeper-3.4.5.1.3.0.0
Note that the command uninstalls HBase. Execute the following command to re-install
HBase:
zypper install hbase-0.94.6.1.3.0.0
Important
When removing and installing packages, rename those files under the /
conf directory that have .rpmsave extension to original to retain the
customized configs. Or, you can also use the configuration files (under the
/conf directory) that you backed up before upgrading.
zypper se -s zookeeper
You should see ZooKeeper v3.4.5.1.3.0.0 in the output. Install ZooKeeper
v3.4.5.1.3.0.0:
zypper install zookeeper-3.4.5.1.3.0.0
2. Start ZooKeeper. On all the ZooKeeper host machines, execute the following command:
sudo su -l $ZOOKEEPER_USER -c "source /etc/zookeeper/conf/zookeeper-env.sh;
export ZOOCFGDIR=/etc/zookeeper/conf; /usr/lib/zookeeper/bin/zkServer.sh
start >> $ZOOKEEPER_LOG_DIR/zoo.out\"

where
$ZOOKEEPER_USER is the ZooKeeper Service user. For example, zookeeper.
$ZOOKEEPER_LOG_DIR is the directory where ZooKeeper server logs are stored. For
example, /var/log/zookeeper.
3. Execute the following commands on the HBase Master and the HBase slave nodes
(RegionServers):
For RHEL/CentOS:
yum upgrade hbase
42
For SLES:
zypper update hbase
4. Start HBase.
a. Start HBase Master. On the HBase Master host machine, execute the following
command:
sudo su -l $HBASE_USER -c "/usr/lib/hbase/bin/hbase-daemon.sh --config /
etc/hbase/conf start master"
b. Start all RegionServers. On all the RegionServers, execute the following command:
sudo su -l $HBASE_USER -c "/usr/lib/hbase/bin/hbase-daemon.sh --config /
etc/hbase/conf start regionserver\"
where $HBASE_USER is the HBase Service user. For example, hbase.
6.4.Upgrade Hive and HCatalog
1. Upgrade Hive and HCatalog. On the Hive and HCatalog host machines, execute the
following commands:
For RHEL/CentOS:
yum upgrade hive hcatalog
For SLES:
zypper update hive hcatalog
yast --update hcatalog hive
2. Start Hive. On the Hive Metastore host machine, execute the following command:
sudo su -l $HIVE_USER -c "nohup hive --service metastore > $HIVE_LOG_DIR/
hive.out 2> $HIVE_LOG_DIR/hive.log &"

3. Start Hive Server2. On the Hive Server2 host machine, execute the following command:
sudo su -l $HIVE_USER -c "nohup /usr/lib/hive/bin/hiveserver2 -hiveconf
hive.metastore.uris=\" \" > $HIVE_LOG_DIR/hiveserver2.out 2> $HIVE_LOG_DIR/
hiveserver2.log &"
where
$HIVE_USER is the Hive Service user. For example, hive.
$HIVE_LOG_DIR is the directory where Hive server logs are stored (example: /var/
log/hive).
6.5.Upgrade Oozie
1. Execute the following command on the Oozie server and client machines:
43
For RHEL/CentOS:
rpm -e --nopostun oozie-$old_version_number
yum install oozie
For SLES:
rpm -e --nopostun oozie-$old_version_number
zypper remove oozie
zypper update oozie-3.3.2.1.3.0.0
zypper se -s oozie
You should see Oozie v3.3.2.1.3.0.0 in the output. Install Oozie v3.3.2.1.3.0.0:
zypper install oozie-3.3.2.1.3.0.0
Important
When removing and installing packages, rename those files the /
conf directory that have .rpmsave extension to original to retain the
customized configs. Alternatively, you can also use the configuration files
(under the /conf directory) you backed up before upgrading.
2. Start Oozie.
sudo su -l $OOZIE_USER -c "cd $OOZIE_LOG_DIR/log; /usr/lib/oozie/bin/oozie-
start.sh"
where
$OOZIE_USER is the Oozie Service user. For example, oozie.
$OOZIE_LOG_DIR is the directory where Oozie log files are stored (for example: /
var/log/oozie).
3. Validate Oozie.
oozie admin -$OOZIE_USER http://$fully.qualified.domain.
name_for_Oozie_hostmachine:11000/oozie -status
You should see the following output:
System mode: NORMAL
6.6.Upgrade WebHCat (Templeton)
1. Remove old Templeton packages. On the Templeton host machine, execute the
following commands:
For RHEL/CentOS:
yum remove templeton\*
For SLES:
zypper remove templeton\*
44
2. Install WebHCat.
For RHEL/CentOS:
yum install webhcat-tar-hive webhcat-tar-pig
For SLES:
zypper install webhcat-tar-hive webhcat-tar-pig
Also see the instructions (provided here) for manually deploying a WebHCat instance.
3. Start WebHCat. On the WebHCat host machine, execute the following command:
sudo su -l $WEBHCAT_USER -c "/usr/lib/hcatalog/sbin/webhcat_server.sh start"

4. Smoke test WebHCat. On the WebHCat host machine, execute the following command:
http://$WEBHCAT_HOST_MACHINE:50111/templeton/v1/status
5. Remove shared libraries from old Templeton installation. On the WebHCat host machine,
execute the following command:
sudo su -l $HDFS_USER -c "hadoop dfs -rmr -skipTrash /apps/templeton"
rm -rf /usr/share/HDP-templeton

where
$WEBHCAT_USER is the WebHCat Service user. For example, hcat.
$HDFS_USER is the HDFS Service user. For example, hdfs.
6.7.Upgrade Pig
1. On all the Pig clients, execute the following command:
For RHEL/CentOS:
yum upgrade pig

For SLES:
zypper update pig

6.8.Upgrade Sqoop
Upgrade Sqoop. On the Sqoop host machine, execute the following command:
For RHEL/CentOS:
yum upgrade sqoop

For SLES:
45
zypper update sqoop

6.9.Upgrade Flume
Upgrade Flume. On the Flume host machine, execute the following command:
For RHEL/CentOS:
yum downgrade flume-1.3.1.1.3.0.0
For SLES:
zypper remove flume
zypper update flume-1.3.1.1.3.0.0
zypper se -s flume
You should see Flume v1.3.1.1.3.0.0 in the output. Install Flume v1.3.1.1.3.0.0:
zypper install flume-1.3.1.1.3.0.0
Important
When removing and installing packages, rename those files the /conf
directory that have .rpmsave extension to original to retain the customized
configs. Alternatively, you can also use the configuration files (under the /
conf directory) you backed up before upgrading.
6.9.1.Validate Flume
By default on installation Flume does not start running immediately. To validate, replace
your default conf/flume.conf with the provdied flume.conf, restart flume, and see if
the data is flowing by examining the destination.
Use this flume.conf file:
1. Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1
2.Describe/configure the source
a1.sources.r1.type = seq
3. Describe the sink
a1.sinks.k1.type = file_roll
a1.sinks.k1.channel = c1
a1.sinks.k1.sink.directory = /tmp/flume
4. Use a channel which buffers events in memory
a1.channels.c1.type = memory
5. Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
After starting flume, check /tmp/flume to see if there are any files there. The files should
contain simple sequential numbers. After validating, stop flume and revert changes to
flume.conf to prevent your disk from filling up.
46
6.10.Upgrade Mahout
Upgrade Mahout. On the Mahout client host machine, execute the following command:
For RHEL/CentOS:
yum upgrade mahout-0.7.0.1.3.0.0
For SLES:
zypper remove mahout
zypper se -s mahout
You should see Mahout v0.7.0.1.3.0.0 in the output. Install Mahout v0.7.0.1.3.0.0:
zypper install mahout-0.7.0.1.3.0.0
Important
When removing and installing packages, rename those files the /conf
directory that have .rpmsave extension to original to retain the customized
configs. Alternatively, you can also use the configuration files (under the /
conf directory) you backed up before upgrading.
6.10.1.Mahout Validation
To validate mahout:
1. Create a test user:
hadoop fs -put /tmp/sample-test.txt /user/testuser
2. Create a mahout test output directory:
hadoop fs -mkdir /user/testuser/mahouttest
3. Set up mahout to convert the plain text file sample-test.txt into a sequence file
that is in the output directory mahouttest.
mahout seqdirectory --input /user/testuser/sample-test.txt --output /user/
testuser/mahouttest --charset utf-8
47
7. Manually Add Slave Nodes to HDP
Cluster
Use the following instructions to manually add slave nodes to your HDP cluster:
1. Prerequisites
2. Add DataNodes or TaskTrackers
3. Add HBase RegionServer
4. Optional - Configure Monitoring Using Ganglia
5. Optional - Configure Cluster Alerting Using Nagios
7.1.Prerequisites
Ensure that the new slave nodes meet the following prerequisites:
The following operating systems are supported:
64-bit Red Hat Enterprise Linux (RHEL) 5 or 6
64-bit CentOS 5 or 6
64-bit SUSE Linux Enterprise Server (SLES) 11, SP1
On each of your hosts:
yum (RHEL)
zypper (SLES)
rpm
scp
curl
wget
unzip
tar
pdsh
Ensure that all the ports listed here are available to the Installer.
To install Hive metastore or to use external database for Oozie metastore, ensure that
you deploy either a MySQL or an Oracle database in your cluster. For instructions, see
here.
48
Your system must have the correct JDK installed on all the nodes of the cluster.
HDP requires Oracle JDK 1.6 update 31. For more information, see Install the Java
Development Kit.
7.2.Add DataNodes or TaskTrackers
Use the following instructions to manually add a DataNode or a TaskTracker hosts:
1. On each of the newly added slave nodes, add the HDP repository to yum:
wget -nv http://dev.hortonworks.com.s3.amazonaws.com/HDP/repos/centos6/hdp.
repo -O
/etc/yum.repos.d/hdp.repo
yum clean all
2. On each of the newly added slave nodes, install HDFS and MapReduce.
On RHEL and CentOS:
yum install hadoop hadoop-libhdfs hadoop-native
yum install hadoop-pipes hadoop-sbin openssl
On SLES:
zypper install hadoop hadoop-libhdfs hadoop-native
zypper install hadoop-pipes hadoop-sbin openssl
3. On each of the newly added slave nodes, install Snappy compression/decompression
library:
a. Check if Snappy is already installed:
rpm-qa | grep snappy
b. Install Snappy on the new nodes:
For RHEL/CentOS:
yum install snappy snappy-devel
For SLES:
zypper install snappy snappy-devel
ln -sf /usr/lib64/libsnappy.so
/usr/lib/hadoop/lib/native/Linux-amd64-64/.
4. Optional - Install the LZO compression library.
On RHEL and CentOS:
yum install lzo-devel hadoop-lzo-native
On SLES:
zypper install lzo-devel hadoop-lzo-native
49
5. Copy the Hadoop configurations to the newly added slave nodes and set appropriate
permissions.
Option I: Copy Hadoop config files from an existing slave node.
a. On an existing slave node, make a copy of the current configurations:
tar zcvf hadoop_conf.tgz /etc/hadoop/conf
b. Copy this file to each of the new nodes:
rm -rf /etc/hadoop/conf
cd /
tar zxvf $location_of_copied_conf_tar_file/hadoop_conf.tgz
chmod -R 755 /etc/hadoop/conf
Option II: Manually add Hadoop configuration files.
a. Download core Hadoop configuration files from here and extract the files under
configuration_files -> core_hadoop directory to a temporary location.
b. In the temporary directory, locate the following files and modify the properties
based on your environment. Search for TODO in the files for the properties to
replace.
Table7.1.core-site.xml
Property Example Description
fs.default.name hdfs://{namenode.full.host-
name}:8020
Enter your NameNode hostname
fs.checkpoint.dir /grid/hadoop/hdfs/snn A comma separated list of paths.
Use the list of directories from
$FS_CHECKPOINT_DIR..
Table7.2.hdfs-site.xml
dfs.name.dir /grid/hadoop/hdfs/nn,/grid1/
hadoop/hdfs/nn
Comma separated list of paths. Use the list
of directories from $DFS_NAME_DIR
dfs.data.dir /grid/hadoop/hdfs/dn,grid1/
hadoop/hdfs/dn
Comma separated list of paths. Use the list
of directories from $DFS_DATA_DIR
dfs.http.address {namenode.full.hostname}:50070 Enter your NameNode hostname for http
access
dfs.secondary.http.address {secondary.namenode.full.host-
name}:50090
Enter your SecondaryNameNode
hostname
dfs.https.address {namenode.full.hostname}:50470 Enter your NameNode hostname for https
access.
Table7.3.mapred-site.xml
mapred.job.tracker {jobtracker.full.host-
name}:50300
Enter your JobTracker hostname
mapred.job.tracker.http.address {jobtracker.full.host-
name}:50030
50
mapred.local.dir /grid/hadoop/mapred,/grid1/
hadoop/mapred
Comma separated list of paths.
$MAPREDUCE_LOCAL_DIR
mapreduce.tasktracker.group hadoop Enter your group. Use the value of
$HADOOP_GROUP
mapreduce.history.server.http.address {jobtracker.full.hostname}
:51111
Table7.4.taskcontroller.cfg
mapred.local.dir /grid/hadoop/mapred,/grid1/
hadoop/mapred
Comma separated list of paths.
$MAPREDUCE_LOCAL_DIR
c. Create the config directory on all hosts in your cluster, copy in all the configuration
files, and set permissions.
rm -r $HADOOP_CONF_DIR
mkdir -p $HADOOP_CONF_DIR
<copy the all the config files to $HADOOP_CONF_DIR>
chmod a+x $HADOOP_CONF_DIR/
chown -R $HDFS_USER:$HADOOP_GROUP $HADOOP_CONF_DIR/../
chmod -R 755 $HADOOP_CONF_DIR/../
where:
$HADOOP_CONF_DIR is the directory for storing the Hadoop configuration files. For
example, /etc/hadoop/conf.
$HDFS_USER is the user owning the HDFS services. For example, hdfs.
$HADOOP_GROUP is a common group shared by services. For example, hadoop.
6. On each of the newly added slave nodes, start HDFS:
su -hdfs
/usr/lib/hadoop/bin/hadoop-daemon.sh --config
$HADOOP_CONF_DIR start datanode
7. On each of the newly added slave nodes, start MapReduce:
su -mapred
/usr/lib/hadoop/bin/hadoop-daemon.sh --config
$HADOOP_CONF_DIR start tasktracker
8. Add new slave nodes.
To add a new NameNode slave (DataNode):
a. On the NameNode host machine, edit the /etc/hadoop/conf/dfs.include
file and add the list of slave nodes' hostnames (separated by newline character).
51
Important
Ensure that you create a new dfs.include file, if the NameNode host
machine does not have an existing copy of this file.
b. On the NameNode host machine, execute the following command:
su hdfs c hadoop dfsadmin refreshNodes
To add a new JobTracker slave (TaskTracker):
a. One the JobTracker host machine, edit the /etc/hadoop/conf/
mapred.include file and add the list of slave nodes' hostnames (separated by
newline character).
Important
Ensure that you create a new mapred.include file, if the JobTracker
host machine does not have an existing copy of this file.
b. On the JobTracker host machine, execute the following command:
su mapred c hadoop mradmin refreshNodes
9. Optional - Enable monitoring on the newly added slave nodes using the instructions
provided here.
10.Optional - Enable cluster alerting on the newly added slave nodes using the instructions
provided here.
7.3.Add HBase RegionServer
Use the following instructions to manually add HBase RegionServer hosts:
1. On each of the newly added slave nodes, install HBase and ZooKeeper.
For RHEL/CentOS:
yum install zookeeper hbase
For SLES:
zypper install zookeeper hbase
2. On each of the newly added slave nodes, add the HDP repository to yum:
wget -nv http://dev.hortonworks.com.s3.amazonaws.com/HDP-1.2.2/repos/
centos6/hdp.repo -O
/etc/yum.repos.d/hdp.repo
yum clean all
3. Copy the HBase configurations to the newly added slave nodes and set appropriate
permissions.
Option I: Copy HBase config files from an existing slave node.
52
a. On any existing slave node, make a copy of the current configurations:
tar zcvf hbase_conf.tgz /etc/hbase/conf
tar zcvf zookeeper_conf.tgz /etc/zookeeper/conf
b. Copy this file to each of the new nodes:
rm -rf /etc/hbase/conf
mkdir -p /etc/hbase/conf
cd /
tar zxvf $location_of_copied_conf_tar_file/hbase_conf.tgz
chmod -R 755 /etc/hbase/conf
rm -rf /etc/zookeeper/conf
mkdir -p /etc/zookeeper/conf
cd /
tar zxvf $location_of_copied_conf_tar_file/zookeeper_conf.tgz
chmod -R 755 /etc/zookeeper/conf
Option II: Manually add Hadoop configuration files.
a. Download the HBase/ZooKeeper config files from here and extract these files
under configuration_files -> hbase and configuration_files ->
zookeeper directories to two temporary locations.
b. Modify the configuration files:
Table7.5.zoo.cfg
Variable Example Description
server.1 $zookeeper.server1.full.hostname
:2888:3888
Enter the 1st ZooKeeper hostname
:2888:3888
Enter the 2nd ZooKeeper hostname
:2888:3888
Enter the 3rd ZooKeeper hostname
Table7.6.hbase-site.xml
Variable Example Description
hbase.rootdir hdfs://$namenode.full.hostname
:8020/apps/hbase/data
Enter the HBase NameNode server
hbase.master.info.bindAddress $hbase.master.full.hostname Enter the HBase Master server hostname
hbase.zookeeper.quorum server1.full.hostname,
server2.full.hostname,
server3.full.hostname
Comma separated list of Zookeeper
servers (match to what is specified in
zoo.cfgbut without port numbers)
4. On all the new slave nodes create the config directory, copy all the config files, and set
the permissions:
rm -r $HBASE_CONF_DIR ;
mkdir -p $HBASE_CONF_DIR ;
copy all the config files to $HBASE_CONF_DIR
53
chmod a+x $HBASE_CONF_DIR/;
chown -R $HBASE_USER:$HADOOP_GROUP $HBASE_CONF_DIR/../ ;
chmod -R 755 $HBASE_CONF_DIR/../
rm -r $ZOOKEEPER_CONF_DIR ;
mkdir -p $ZOOKEEPER_CONF_DIR ;
copy all the config files to $ZOOKEEPER_CONF_DIR
chmod a+x $ZOOKEEPER_CONF_DIR/;
chown -R $ZOOKEEPER_USER:$HADOOP_GROUP $ZOOKEEPER_CONF_DIR/../ ;
chmod -R 755 $ZOOKEEPER_CONF_DIR/../

where:
$HBASE_CONF_DIR is the directory to store the HBase configuration files. For
example, /etc/hbase/conf.
$HBASE_USER is the user owning the HBase services. For example, hbase.
$HADOOP_GROUP is a common group shared by services. For example, hadoop.
$ZOOKEEPER_CONF_DIR is the directory to store the ZooKeeper configuration files.
For example, /etc/zookeeper/conf
$ZOOKEEPER_USER is the user owning the ZooKeeper services. For example,
zookeeper.
5. Start HBase RegionServer node:
<login as $HBASE_USER>
/usr/lib/hbase/bin/hbase-daemon.sh --config $HBASE_CONF_DIR start
regionserver
6. On the HBase Master host machine, edit the /usr/lib/hbase/conf file and add the
list of slave nodes' hostnames. The hostnames must be separated by a newline character.
7. Optional: Enable monitoring on the newly added slave nodes using the instructions
provided here.
8. Optional: Enable cluster alerting on the newly added slave nodes using the instructions
provided here.
7.4.Optional - Configure Monitoring Using
Ganglia
For each of the newly added slave node, complete the following instructions in order to use
Ganglia for monitoring and metrics collection:
1. On each new host, install the Ganglia gmond:
On RHEL and CentOS:
yum install ganglia-gmond-3.2.0-99
54
On SLES:
zypper install ganglia-gmond-3.2.0-99
2. Copy the HDP Ganglia scripts from any existing slave node to the newly added node:
mkdir -p /usr/libexec/hdp/ganglia
scp root@$FQDN_of_any_existing_node:/usr/libexec/hdp/ganglia/*
/usr/libexec/hdp/ganglia
3. Configure the gmond emitter:
/usr/libexec/hdp/ganglia/setupGanglia.sh -c HDPSlaves
4. Edit the /etc/ganglia/hdp/HDPSlaves/conf.d/gmond.slave.confto validate
the slave configuration files.
Ensure that host points to the Ganglia server master node as shown below:
host = $Ganglia_server_hostname
5. Copy Ganglia service start file from any existing slave node to each of the newly added
node:
scp root@$existing_node:/etc/init.d/hdp-gmond /etc/init.d
/etc/init.d/hdp-gmond start
This step will start Ganglia monitoring for your Hadoop cluster.
7.5.Optional - Configure Cluster Alerting Using
Nagios
For each of the newly added slave node, complete the following instructions in order to use
Nagios for cluster alerting:
1. On the Nagios server, add the host definition to the /etc/nagios/objects/
hadoop-hosts.cfg file:
2. On the Nagios server, edit the /etc/nagios/objects/hadoop-hostsgroups.cfg
file.
Add hostnames of the newly added slave nodes to members list under the slaves host
group.
3. Restart the Nagios server:
service nagios start
55
8. Decommission Slave Nodes
Hadoop provides the decommission feature to retire a set of existing slave nodes
(DataNodes, TaskTrackers, or HBase RegionServers) in order to prevent data loss.
Slaves nodes are frequently decommissioned for maintainance. As a Hadoop administrator,
you will decommission the slave nodes periodically in order to either reduce the cluster size
or to gracefully remove dying nodes.
Use the following sections to decommission slave nodes in your cluster:
Prerequisites
Decommission DataNodes or TaskTrackers
Decommission HBase RegionServers
8.1.Prerequisites
Ensure that the following property is defined in your hdfs-site.xml file.
<property>
<name>dfs.hosts.exclude</name>
<value>$HADOOP_CONF_DIR/excludes</value>
<final>true</final>
</property>

Ensure that the following property is defined in your mapred-site.xml file.
<property>
<name>mapred.hosts.exclude </name>
<value>$HADOOP_CONF_DIR/excludes</value>
<final>true</final>
</property>

8.2.Decommission DataNodes or TaskTrackers
Nodes normally run both a DataNode and a TaskTracker, and both are typically
commissioned or decommissioned together.
With the replication level set to three, HDFS is resilient to individual DataNodes failures.
However, there is a high chance of data loss when you terminate DataNodes without
decommissioning them first. Nodes must be decommissioned on a schedule that permits
replication of blocks being decommissioned.
56
On the other hand, if a TaskTracker is shutdown, the JobTracker will schedule the tasks
on other TaskTrackers. However, decommissioning a TaskTracker is required especially in
situations where you want that TaskTracker to stop to accepting new tasks or when the
tasks take time to execute but you still want to be agile in your cluster management.
8.2.1.Decommission DataNodes
Use the following instructions to decommission DataNodes in your cluster:
1. On the NameNode host machine, edit the $HADOOP_CONF_DIR/dfs.exclude file
and add the list of DataNodes hostnames (separated by newline character).
2. Update the NameNode with the new set of permitted DataNodes. On the NameNode
host machine, execute the following command:
su - $HDFS_USER
% hadoop dfsadmin refreshNodes
where $HDFS_USER is the user owning the HDFS services. For example, hdfs.
3. Open the NameNode web UI (http://$NameNode_FQDN:50070) and navigate to
Decommissioning Nodes page.
Check whether the admin state has changed to Decommission In Progress for the
DataNodes being decommissioned.
4. When all the DataNodes report their state as Decommissioned then all the blocks have
been replicated. Shut down the decommissioned nodes.
5. On the NameNode host machine, remove the decommissioned nodes from the
$HADOOP_CONF_DIR/dfs.include file and execute the following command:
su - $HDFS_USER
% hadoop dfsadmin refreshNodes
8.2.2.Decommission TaskTrackers
Use the following instructions to decommission TaskTrackers in your cluster:
1. On the JobTracker host machine, edit the $HADOOP_CONF_DIR/mapred.exclude file
and add the list of TaskTrackers hostnames (separated by newline character).
2. Update the JobTracker with the new set of permitted TaskTrackers. On the JobTracker
host machine, execute the following command:
su - $MAPRED_USER
% hadoop mradmin refreshNodes
where $MAPRED_USER is the user owning the MapReduce services. For example,
mapred.
57
8.3.Decommission HBase RegionServers
Use the following instructions to decommission HBase RegionServers in your cluster:
1. Decommission RegionServers.
The preferred method of decommissioning RegionServers is to use the
graceful_stop.sh script (Option I).This option gradually unloads Regions from the
RegionServer, allowing the node to be terminated without impacting data availability.
You can also terminate the RegionServer without first unloading its Regions (Option II).
This will result in a short window of data unavailability as HBase's natural data recovery
operations execute.
Option I: Perform graceful stop
You can also use the following command to gracefully decommission a loaded
RegionServer.
Execute the following command from any host machine with HBase configuration
installed:
su - $HBASE_USER
/usr/lib/hbase/bin/graceful_stop.sh $RegionServer.Hostname
where $HBASE_USER is the user owning the HBase Services. For example, hbase.
Important
The value of $RegionServer.Hostname argument must match the
hostname that HBase uses to identify RegionServers.
To find the hostname for a particular RegionServer, go to the HBase web
UI and check the list of RegionServers in the HBase master UI. Typically,
HBase Master uses hostnames but occassionally it can be the FQDN of a
RegionServer.
Option II: Use hbase-daemon.sh
Execute the following command on the RegionServer that you want to decommission:
Important
It is important to execute the hbase-daemon.sh script on the
RegionServer that you want to decommission.
su - $HBASE_USER
/usr/lib/hbase/bin/hbase-daemon.sh stop regionserver
where $HBASE_USER is the user owning the HBase Services. For example, hbase.
Note that Option II causes the RegionServer to close all the regions before shutting
down.
2. Enable the load balancer.
58
If you used the graceful_stop.sh script earlier, you may need to re-enable the
Region Balancer. Do so using the balance_switch command from the shell. Pass
the command true to enable the balancer, false to disable it. The command's return
value is the state of the balancer before running the command. If graceful_stop.sh
disabled the balancer earlier, enable it again like this:
su - $HBASE_USER
hbase shell
hbase(main):001:0> balance_switch true
false
0 row(s) in 0.3590 seconds
where $HBASE_USER is the user owning the HBase services. For example, hbase
59
9. HDP Logfiles
The following sections detail HDP logging and Backups:
1. HDP Log Locations
2. HDP Log Format
3. HDP Backups
9.1.HDP Log Locations
Each component of the HDP ecosystem creates logs from each of the daemons under
normal operation, as well as configuration logs, statistics, standard error, standard out, and
internal diagnostic information.
The following table for HDP 1.3 lists the types of logs and where you can find them. The
locations depend on the user that owns the process. For example, $hdfs represents the
system user that owns the HDFS component daemons.
Table9.1.Dependencies on HDP components
Component Log Type Log Location (defaults)
HDFS NameNode operational logs On NameNode - /var/log/hadoop/
$hdfs
HDFS DataNode operational logs On DataNodes - /var/log/hadoop/
$hdfs
MapReduce (HDP 1.x) Job Tracker operational logs On JobTracker - /var/log/hadoop/
$mapred
MapReduce (HDP 1.x) Task Tracker operational logs On TaskTrackers - /var/log/
hadoop/$mapred
MapReduce (HDP 1.x) MapReduce Job logs On TaskTrackers - /var/log/
hadoop/$mapred/userlogs
Hive HiveServer2 operational logs On HiveServer2 - /var/log/hive/
hive-server2.*
Hive Hive Metastore operational logs On Hive Metastore - /var/log/
hive/hive.*
WebHCat WebHCat Server operational logs On WebHCat server - /var/log/
webhcat
Oozie Oozie Server operational logs On Oozie Server - /var/log/oozie
HBase HBase Master operational logs On HBase Master - /var/log/hbase
HBase Region Server operational logs On RegionServers - /var/log/hbase
ZooKeeper ZooKeeper operational logs On ZooKeeper nodes - /var/log/
zookeeper
Nagios Nagios Server operational logs On the Nagios Server - /var/log/
nagios Log Archive Path - /var/
log/nagios/archives
Ganglia Ganglia Server On the Ganglia Server - /var/log/
messages
Ganglia Ganglia Monitor On Ganglia Monitor nodes - /var/
log/messages
60
Component Log Type Log Location (defaults)
Ambari Ambari Server On Ambari Server - /var/log/
ambari-server
Ambari Ambari Agent On Ambari Server - /var/log/
ambari-server
9.2.HDP Log Format
Hadoop and the other HDP components use Log4J to format their log files, and to
filter what information is logged. The log format and filter can be configured for each
component by editing the corresponding log4j.properties file. These files are found
in the configuration folders for each component. The following log4j severity levels are
available to filter logs:
OFF- Highest possible rank. Turns off all logging.
FATAL - Application will probably abort due to these errors. Displays on the status
console.
ERROR - Runtime errors or unexpected conditions. Application may continue to run.
Displays on console.
WARN - Undesirable or unexpected conditions. Potentially danerous conditions. Displays
on console.
INFO - Interesting messages on application progress. Displays on console.
DEBUG - Detailed information on events to debug an application. Written to logs only.
TRACE - More detailed information than DEBUG to help debug an application. Written to
logs only.
ALL - Lowest possible rank. Turns on all logging.
Note
If you do not assign a logging level to the logger, the logger inherits the first
non-null level in the hirearchy. By default, the root logger is at the DEBUG
logging level.
9.3.HDP Backups
You can back up your HDP clusters to an external backup HDP cluster using distcp, a bulk
data movement tool. Invoke distcp periodically to transfer HDFS datasets from the active
HDP cluster to the backup HDP cluster.
For more information on backing up clusters with distcp, visit the distcp community page
here.
61
10. WebHDFS Administrator Guide
Use the following instructions to set up WebHDFS:
1. Set up WebHDFS.
Add the following property to the hdfs-site.xml file:
<property>
<name>dfs.webhdfs.enabled</name>
<value>true</value>
</property>
2. [Optional] - If running a secure cluster, follow the steps listed below.
a. Create an HTTP service user principal using the command given below:
kadmin: addprinc -randkey HTTP/$<Fully_Qualified_Domain_Name>@
$<Realm_Name>.COM
where:
Fully_Qualified_Domain_Name: Host where NameNode is deployed
Realm_Name: Name of your Kerberos realm
b. Create keytab files for the HTTP principals.
kadmin: xst -norandkey -k /etc/security/spnego.service.keytab HTTP/
$<Fully_Qualified_Domain_Name>
c. Verify that the keytab file and the principal are associated with the correct service.
klist k -t /etc/security/spnego.service.keytab
d. Add the following properties to the hdfs-site.xml file.
<property>
<name>dfs.web.authentication.kerberos.principal</name>
<value>HTTP/$<Fully_Qualified_Domain_Name>@$<Realm_Name>.COM</value>
</property>
<property>
<name>dfs.web.authentication.kerberos.keytab</name>
<value>/etc/security/spnego.service.keytab</value>
</property>
where:
Fully_Qualified_Domain_Name: Host where NameNode is deployed
Realm_Name: Name of your Kerberos realm
62
3. Restart the NameNode and DataNode services using the instructions provided here.

Hortonworks Hadoop System Admin Guide 20130819

Uploaded by

Copyright:

Available Formats

Hortonworks Hadoop System Admin Guide 20130819

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hortonworks Hadoop System Admin Guide 20130819

Uploaded by

Copyright:

Available Formats

What is Hortonworks Data Platform?

What components does HDP consist of?

docs.hortonworks.

You might also like