Posts Tagged ‘Big Data’

Steps to Configure a Single-Node YARN Cluster

March 30, 2016 Leave a comment

The following type of installation is often referred to as “pseudo-distributed” because it mimics some of the functionality of a distributed Hadoop cluster. A single machine is, of course, not practical for any production use, nor is it parallel. A small-scale Hadoop installation can provide a simple method for learning Hadoop basics, however.

The recommended minimal installation hardware is a dual-core processor with 2 GB of RAM and 2 GB of available hard drive space. The system will need a recent Linux distribution with Java installed (e.g., Red Hat Enterprise Linux or rebuilds, Fedora, Suse Linux Enterprise, OpenSuse, Ubuntu). Red Hat Enterprise Linux 6.3 is used for this installation example. A bash shell environment is also assumed. The first step is to download Apache Hadoop.

Step 1: Download Apache Hadoop

Download the latest distribution from the Hadoop website ( http://hadoop.apache. org/). For example, as root do the following:

# cd /root
# wget ➥2.2.0.tar.gz

Next create and extract the package in /opt/yarn:

# mkdir –p /opt/yarn

# cd /opt/yarn
# tar xvzf /root/hadoop-2.2.0.tar.gz

Step 2: Set JAVA_HOME

For Hadoop 2, the recommended version of Java can be found at http://wiki.apache. org/hadoop/HadoopJavaVersions. In general, a Java Development Kit 1.6 (or greater) should work. For this install, we will use Open Java 1.6.0_24, which is part of Red Hat Enterprise Linux 6.3. Make sure you have a working Java JDK installed; in this case, it is the Java-1.6.0-openjdk RPM. To include JAVA_HOME for all bash users (other shells must be set in a similar fashion), make an entry in /etc/profile.d as follows:

# echo “export JAVA_HOME=/usr/lib/jvm/java-1.6.0-openjdk-” > /etc/ ➥profile.d/

To make sure JAVA_HOME is defined for this session, source the new script: # source /etc/profile.d/

Step 3: Create Users and Groups

It is best to run the various daemons with separate accounts. Three accounts (yarn, hdfs, mapred) in the group hadoop can be created as follows:

# groupadd hadoop

# useradd -g hadoop yarn

# useradd -g hadoop hdfs

# useradd -g hadoop mapred

Step 4: Make Data and Log Directories

Hadoop needs various data and log directories with various permissions. Enter the following lines to create these directories:

# mkdir -p /var/data/hadoop/hdfs/nn

# mkdir -p /var/data/hadoop/hdfs/snn

# mkdir -p /var/data/hadoop/hdfs/dn

# chown hdfs:hadoop /var/data/hadoop/hdfs –R

# mkdir -p /var/log/hadoop/yarn

# chown yarn:hadoop /var/log/hadoop/yarn -R

Next, move to the YARN installation root and create the log directory and set the owner and group as follows:

# cd /opt/yarn/hadoop-2.2.0

# mkdir logs
# chmod g+w logs

# chown yarn:hadoop . -R

Step 5: Configure core-site.xml

From the base of the Hadoop installation path (e.g., /opt/yarn/hadoop-2.2.0), edit the etc/hadoop/core-site.xml file. The original installed file will have no entries other than the<configuration></configuration>tags. Two properties need to be set. The first is the property, which sets the host and request port name for the NameNode (metadata server for HDFS). The second is hadoop.http.staticuser.user, which will set the default user name to hdfs. Copy the following lines to the Hadoop etc/hadoop/core-site.xml file and remove the original empty <configuration> </configuration> tags.







<name>hadoop.http.staticuser.user</name> <value>hdfs</value>



Step 6: Configure hdfs-site.xml

From the base of the Hadoop installation path, edit the etc/hadoop/hdfs-site.xml file. In the single-node pseudo-distributed mode, we don’t need or want the HDFS to replicate file blocks. By default, HDFS keeps three copies of each file in the file system for redundancy. There is no need for replication on a single machine; thus the value of dfs.replication will be set to 1.

In hdfs-site.xml, we specify the NameNode, Secondary NameNode, and Data- Node data directories that we created in Step 4. These are the directories used by the various components of HDFS to store data. Copy the following lines into Hadoop etc/hadoop/hdfs-site.xml and remove the original empty <configuration> </configuration> tags.




<value>1</value> </property> <property>


<value>file:/var/data/hadoop/hdfs/nn</value> </property>






<name>fs.checkpoint.edits.dir</name> <value>file:/var/data/hadoop/hdfs/snn</value>








Step 7: Configure mapred-site.xml

From the base of the Hadoop installation, edit the etc/hadoop/mapred-site.xml file. A new configuration option for Hadoop 2 is the capability to specify a framework name for MapReduce, setting the property. In this install, we will use the value of “yarn” to tell MapReduce that it will run as a YARN appli- cation. First, copy the template file to the mapred-site.xml.

# cp mapred-site.xml.template mapred-site.xml

Next, copy the following lines into Hadoop etc/hadoop/mapred-site.xml file and

remove the original empty <configuration> </configuration> tags.








Step 8: Configure yarn-site.xml

From the base of the Hadoop installation, edit the etc/hadoop/yarn-site.xml file. The yarn.nodemanager.aux-services property tells NodeManagers that there will be an auxiliary service called mapreduce.shuffle that they need to implement. After we tell the NodeManagers to implement that service, we give it a class name as the means to implement that service. This particular configuration tells MapReduce how to do its shuffle. Because NodeManagers won’t shuffle data for a non-MapReduce job by default, we need to configure such a service for MapReduce. Copy the following lines to the Hadoop etc/hadoop/yarn-site.xml file and remove the original empty <configuration> </configuration> tags.








<value>org.apache.hadoop.mapred.ShuffleHandler</value> </property>


Step 9: Modify Java Heap Sizes

The Hadoop installation uses several environment variables that determine the heap sizes for each Hadoop process. These are defined in the etc/hadoop/* files used by Hadoop. The default for most of the processes is a 1 GB heap size; because we’re running on a workstation that will probably have limited resources compared to a standard server, however, we need to adjust the heap size settings. The values that follow are adequate for a small workstation or server.

Edit the etc/hadoop/ file to reflect the following (don’t forget to remove the “#” at the beginning of the line):



Next, edit to ref lect the following:


Finally, edit to ref lect the following:


The following line will need to be added to


Step 10: Format HDFS

For the HDFS NameNode to start, it needs to initialize the directory where it will hold its data. The NameNode service tracks all the metadata for the file sys- tem. The format process will use the value assigned to in etc/hadoop/hdfs-site.xml earlier (i.e., /var/data/hadoop/hdfs/nn). Format- ting destroys everything in the directory and sets up a new file system. Format the NameNode directory as the HDFS superuser, which is typically the “hdfs” user account.

From the base of the Hadoop distribution, change directories to the “bin” direc- tory and execute the following commands:

# su – hdfs

$ cd /opt/yarn/hadoop-2.2.0/bin

$ ./hdfs namenode -format

If the command worked, you should see the following near the end of a long list of messages:

INFO common.Storage: Storage directory /var/data/hadoop/hdfs/nn has been ➥successfully formatted.


Step 11: Start the HDFS Services

Once formatting is successful, the HDFS services must be started. There is one ser- vice for the NameNode (metadata server), a single DataNode (where the actual data
is stored), and the SecondaryNameNode (checkpoint data for the NameNode). The Hadoop distribution includes scripts that set up these commands as well as name other values such as PID directories, log directories, and other standard process configura- tions. From the bin directory in Step 10, execute the following as user hdfs:

$ cd ../sbin

$ ./ start namenode

The command should show the following:

starting namenode, logging to /opt/yarn/hadoop-2.2.0/logs/hadoop-hdfs-namenode- ➥limulus.out

The secondarynamenode and datanode services can be started in the same way:

$ ./ start secondarynamenode
starting secondarynamenode, logging to /opt/yarn/hadoop-2.2.0/logs/hadoop-hdfs- ➥secondarynamenode-limulus.out

$ ./ start datanode
starting datanode, logging to /opt/yarn/hadoop-2.2.0/logs/hadoop-hdfs-datanode- ➥limulus.out

If the daemon started successfully, you should see responses that will point to the log file. (Note that the actual log file is appended with “.log,” not “.out.”). As a sanity check, issue a jps command to confirm that all the services are running. The actual PID (Java Process ID) values will be different than shown in this listing:

$ jps

15140 SecondaryNameNode

15015 NameNode

15335 Jps

15214 DataNode

If the process did not start, it may be helpful to inspect the log files. For instance, examine the log file for the NameNode. (Note that the path is taken from the preced- ing command.)

vi /opt/yarn/hadoop-2.2.0/logs/hadoop-hdfs-namenode-limulus.log

All Hadoop services can be stopped using the script. For example, to stop the datanode service, enter the following (as user hdfs in the /opt/yarn/hadoop-2.2.0/sbin directory):

$ ./ stop datanode

The same can be done for the NameNode and SecondaryNameNode.

Step 12: Start YARN Services

As with HDFS services, the YARN services need to be started. One ResourceManager and one NodeManager must be started as user yarn (exiting from user hdfs first):

$ exit


# su – yarn

$ cd /opt/yarn/hadoop-2.2.0/sbin

$ ./ start resourcemanager

starting resourcemanager, logging to /opt/yarn/hadoop-2.2.0/logs/yarn-yarn- ➥resourcemanager-limulus.out

$ ./ start nodemanager
starting nodemanager, logging to /opt/yarn/hadoop-2.2.0/logs/yarn-yarn- ➥nodemanager-limulus.out

As when the HDFS daemons were started in Step 1, the status of the running dae- mons is sent to their respective log files. To check whether the services are running, issue a jps command. The following shows all the services necessary to run YARN on a single server:

$ jps

15933 Jps

15567 ResourceManager

15785 NodeManager

If there are missing services, check the log file for the specific service. Similar to the case with HDFS services, the services can be stopped by issuing a stop argument to the daemon script:

./ stop nodemanager

Step 13: Verify the Running Services Using the Web Interface

Both HDFS and the YARN ResourceManager have a web interface. These interfaces are a convenient way to browse many of the aspects of your Hadoop installation. To monitor HDFS, enter the following (or use your favorite web browser):

$ firefox http://localhost:50070

Connecting to port 50070 will bring up a web interface similar to Figure 1.
1 .Web interface for the ResourceManager can be viewed by entering the following:

$ firefox http://localhost:8088

A webpage similar to that shown in Figure 1.2 will be displayed.

Ekran Resmi 2016-03-31 01.25.42

Figure 1.1 Webpage for HDFS file system


Ekran Resmi 2016-03-31 01.25.34

Figure 1.2 Webpage for YARN ResourceManager


How-to: Quickly Configure Kerberos for Your Apache Hadoop Cluster “”

March 11, 2016 Leave a comment

Use the scripts and screenshots below to configure a Kerberized cluster in minutes.

Kerberos is the foundation of securing your Apache Hadoop cluster. With Kerberos enabled, user authentication is required. Once users are authenticated, you can use projects like Apache Sentry (incubating) for role-based access control via GRANT/REVOKE statements.

Taming the three-headed dog that guards the gates of Hades is challenging, so Cloudera has put significant effort into making this process easier in Hadoop-based enterprise data hubs. In this post, you’ll learn how to stand-up a one-node cluster with Kerberos enforcing user authentication, using the Cloudera QuickStart VM as a demo environment.

If you want to read the product documentation, it’s available here. You should consider this reference material; I’d suggest reading it later to understand more details about what the scripts do.


You need the following downloads to follow along.

Initial Configuration

Before you start the QuickStart VM, increase the memory allocation to 8GB RAM and increase the number of CPUs to two. You can get by with a little less RAM, but we will have everything including the Kerberos server running on one node.

Start up the VM and activate Cloudera Manager as shown here:

Give this script some time to run, it has to restart the cluster.

KDC Install and Setup Script

The script does all the setup work for the Kerberos server and the appropriate configuration parameters. The comments are designed to explain what is going on inline. (Do not copy and paste this script! It contains unprintable characters that are pretending to be spaces. Rather, download it.)

Cloudera Manager Kerberos Wizard

After running the script, you now have a working Kerberos server and can secure the Hadoop cluster. The wizard will do most of the heavy lifting; you just have to fill in a few values.

To start, log into Cloudera Manager by going to http://quickstart.cloudera:7180 in your browser. The userid is cloudera and the password is cloudera. (Almost needless to say but never use “cloudera” as a password in a real-world setting.)

There are lots of productivity tools here for managing the cluster but ignore them for now and head straight for the Administration > Kerberos wizard as shown in the next screenshot.

Click on the “Enable Kerberos” button.

The four checklist items were all completed by the script you’ve already run. Check off each item and select “Continue.”

The Kerberos Wizard needs to know the details of what the script configured. Fill in the entries as follows:

  • KDC Server Host: quickstart.cloudera
  • Kerberos Security Realm: CLOUDERA
  • Kerberos Encryption Types: aes256-cts-hmac-sha1-96

Click “Continue.”

Do you want Cloudera Manager to manage the krb5.conf files in your cluster? Remember, the whole point of this blog post is to make Kerberos easier. So, please check “Yes” and then select “Continue.”

The Kerberos Wizard is going to create Kerberos principals for the different services in the cluster. To do that it needs a Kerberos Administrator ID. The ID created is: cloudera-scm/admin@CLOUDERA.

The screen shot shows how to enter this information. Recall the password is: cloudera.

The next screen provides good news. It lets you know that the wizard was able to successfully authenticate.

OK, you’re ready to let the Kerberos Wizard do its work. Since this is a VM, you can safely select “I’m ready to restart the cluster now” and then click “Continue.” You now have time to go get a coffee or other beverage of your choice.

How long does that take? Just let it work.

Congrats, you are now running a Hadoop cluster secured with Kerberos.

Kerberos is Enabled. Now What?

The old method of su - hdfs will no longer provide administrator access to the HDFS filesystem. Here is how you become the hdfs user with Kerberos:

Now validate you can do hdfs user things:

Next, invalidate the Kerberos token so as not to break anything:

The min.user parameter needs to be fixed per the message below:

This is the error message you get without fixing

Save the changes shown above and restart the YARN service. Now validate that the cloudera user can use the cluster:

If you forget to kinit before trying to use the cluster you’ll get the errors below. The simple fix is to use kinit with the principal you wish to use.

Acquiring Big Data Using Apache Flume

December 29, 2015 Leave a comment
No technology is more synonymous with Big Data than Apache Hadoop. Hadoop’s distributed filesystem and compute framework make possible cost-effective, linearly scalable processing of petabytes of data. Unfortunately, there are few tutorials devoted to how to get big data into Hadoop in the first place.
Some data destined for Hadoop clusters surely comes from sporadic bulk loading processes, such as database and mainframe offloads and batched data dumps from legacy systems. But what has made data really big in recent years is that most new data is contained in high-throughput streams. Application logs, GPS tracking, social media updates, and digital sensors all constitute fast-moving streams begging for storage in the Hadoop Distributed File System (HDFS). As you might expect, several technologies have been developed to address the need for collection and transport of these high-throughput streams. Facebook’s Scribe and Apache/LinkedIn’s Kafka both offer solutions to the problem, but Apache Flume is rapidly becoming a de facto standard for directing data streams into Hadoop.

This article describes the basics of Apache Flume and illustrates how to quickly set up Flume agents for collecting fast-moving data streams and pushing the data into Hadoop’s filesystem. By the time we’re finished, you should be able to configure and launch a Flume agent and understand how multi-hop and fan-out flows are easily constructed from multiple agents.

Anatomy of a Flume Agent

Flume deploys as one or more agents, each contained within its own instance of the Java Virtual Machine (JVM). Agents consist of three pluggable components: sources, sinks, and channels. An agent must have at least one of each in order to run. Sources collect incoming data as events. Sinks write events out, and channels provide a queue to connect the source and sink. (Figure 1.)

Apache Flume Agent
Figure 1: Flume Agents consist of sources, channels, and sinks.


Put simply, Flume sources listen for and consume events. Events can range from newline-terminated strings in stdout to HTTP POSTs and RPC calls — it all depends on what sources the agent is configured to use. Flume agents may have more than one source, but must have at least one. Sources require a name and a type; the type then dictates additional configuration parameters.

On consuming an event, Flume sources write the event to a channel. Importantly, sources write to their channels as transactions. By dealing in events and transactions, Flume agents maintain end-to-end flow reliability. Events are not dropped inside a Flume agent unless the channel is explicitly allowed to discard them due to a full queue.


Channels are the mechanism by which Flume agents transfer events from their sources to their sinks. Events written to the channel by a source are not removed from the channel until a sink removes that event in a transaction. This allows Flume sinks to retry writes in the event of a failure in the external repository (such as HDFS or an outgoing network connection). For example, if the network between a Flume agent and a Hadoop cluster goes down, the channel will keep all events queued until the sink can correctly write to the cluster and close its transactions with the channel.

Channels are typically of two types: in-memory queues and durable disk-backed queues. In-memory channels provide high throughput but no recovery if an agent fails. File or database-backed channels, on the other hand, are durable. They support full recovery and event replay in the case of agent failure.


Sinks provide Flume agents pluggable output capability — if you need to write to a new type storage, just write a Java class that implements the necessary classes. Like sources, sinks correspond to a type of output: writes to HDFS or HBase, remote procedure calls to other agents, or any number of other external repositories. Sinks remove events from the channel in transactions and write them to output. Transactions close when the event is successfully written, ensuring that all events are committed to their final destination.

Setting up a Simple Agent for HDFS

A simple one-source, one-sink Flume agent can be configured with just a single configuration file. In this example, I’ll create a text file named sample_agent.conf — it looks a lot like a Java properties file. At the top of the file, I configure the agent’s name and the names of its source, sink, and channel.

hdfs-agent.sources= netcat-collect
hdfs-agent.sinks = hdfs-write
hdfs-agent.channels= memory-channel

This defines an agent named hdfs-agent and the names of the sources, sinks, and channels; keep the name in mind, because we’ll need it to start the agent. Multiple sources, sinks, and channels can be defined on these lines as a whitespace-delimited list of names. In this case, the source is named netcat-collect, the sink hdfs-write, and the channel is named memory-channel. The names are indicative of what I’m setting up: events collected via netcat will be written to HDFS and I will use a memory-only queue for transactions.

Next, I configure the source. I use a netcat source, as it provides a simple means of interactively testing the agent. A netcat source requires a type as well as an address and port to which it should bind. The netcat source will listen on localhost on port 11111; messages to netcat will be consumed by the source as events.


hdfs-agent.sources.netcat-collect.type = netcat
hdfs-agent.sources.netcat-collect.bind =
hdfs-agent.sources.netcat-collect.port = 11111

With the source defined, I’ll configure the sink to write to HDFS. HDFS sinks support a number of options, but by default, the HDFS sink writes Hadoop SequenceFiles. In this example, I’ll specify the sink write raw textfiles to HDFS so they can be easily inspected; I’ll also set a roll interval, which forces Flume to commit writes to HDFS every 30 seconds. File rolls can be configured based on time, size, or a combination of the two. The file rolls are particularly important in environments for which HDFS does not support appending to files.

hdfs-agent.sinks.hdfs-write.type = hdfs
hdfs-agent.sinks.hdfs-write.hdfs.path = hdfs://namenode_address:8020/path/to/flume_test
hdfs-agent.sinks.hdfs-write.rollInterval = 30

Finally, I’ll configure a memory-backed channel to transfer events from source to sink and connect them together. Keep in mind that if I exceed the channel capacity, Flume will drop events. If I need durability in a file or JDBC, channel should be used instead.


hdfs-agent.channels.memoryChannel.type = memory

With the configuration complete, I start the Flume agent from a terminal:

flume-ng agent -f /path/to/sample_agent.conf -n hdfs-agent

In the Flume agent’s logs, I look for indication that the source, sink, and channel have successfully started. For example:

INFO nodemanager.DefaultLogicalNodeManager: Starting Channel memoryChannel
INFO instrumentation.MonitoredCounterGroup: Component type: CHANNEL, name: memoryChannel started
INFO nodemanager.DefaultLogicalNodeManager: Starting Sink hdfs-write
INFO nodemanager.DefaultLogicalNodeManager: Starting Source netcat-collect
INFO source.NetcatSource: Source starting

In a separate terminal, connect to the agent via netcat and enter a series of messages.

> nc localhost 11111
> testing
> 1
> 2
> 3
In the agent logs, an HDFS file will be created and committed every 30 seconds. If I print the contents of the files to standard out using HDFS cat, I’ll find the messages from netcat are stored.

More Advanced Deployments

Regardless of source, direct writers to HDFS are too simple to be suitable for many deployments: Application servers may reside in the cloud while clusters are on-premise, many streams of data may need to be consolidated, or events may need to be filtered during transmission. Fortunately, Flume easily enables reliable multi-hop event transmission. Fan-in and fan-out patterns are readily supported via multiple sources and channel options. Additionally, Flume provides the notion of interceptors, which allow the decoration and filtering of events in flight.

Multi-Hop Topologies

Flume provides multi-hop deployments via Apache Avro-serialized RPC calls. For a given hop, the sending agent implements an Avro sink directed to a host and port where the receiving agent is listening. The receiver implements an Avro source bound to the designated host-port combination. Reliability is ensured by Flume’s transaction model. The sink on the sending agent does not close its transaction until receipt is acknowledged by the receiver. Similarly, the receiver does not acknowledge receipt until the incoming event has been committed to its channel.

#sender configuration
avro-agent.sinks= avro-sink
#receiver configuration on

Multihop event flow
Figure 2: Multihop event flows are constructed using RPCs between Avro sources and sinks.

Fan-In and Fan-Out

Fan-in is a common case for Flume agents. Agents may be run on many data collectors (such as application servers) in a large deployment, while only one or two writers to a remote Hadoop cluster are required to handle the total event throughput. In this case, the Flume topology is simple to configure. Each agent at a data collector implements the appropriate source and an Avro sink. All Avro sinks point to the host and port of the Flume agent charged with writing to the Hadoop cluster. The agent at the Hadoop cluster simply configures an Avro source on the designated host and port. Incoming events are consolidated automatically and are written to the configured sink.

Fan-out topologies are enabled via Flume’s source selectors. Selectors can be replicating — sending all events to multiple channels — or multiplexing. Multiplexed sources can be partitioned by mappings defined on events via interceptors. For example, a replicating selector may be appropriate when events need to be sent to HDFS and to flat log files or a database. Multiplexed selectors are useful when different mappings should be directed to different writers; for example, data destined for partitioned Hive tables may be best handled via multiplexing.

hdfs-agent.channels=mchannel1 mchannel2
hdfs-agent.sources.netcat-collect.selector.type = replicating
hdfs-agent.sources.r1.channels = mchannel1 mchannel2


Flume provides a robust system of interceptors for in-flight modification of events. Some interceptors serve to decorate data with metadata useful in multiplexing the data or processing it after it has been written to the sink. Common decorations include timestamps, hostnames, and static headers. It’s a great way to keep track of when your data arrived and from where it came.

More interestingly, interceptors can be used to selectively filter or decorate events. The Regex Filtering Interceptor allows events to be dropped if they match the provided regular expression. Similarly, the Regex Extractor Interceptor decorates event headers according to a regular expression. This is useful if incoming events require multiplexing, but static definitions are too inflexible.

hdfs-agent.sources.netcat-collect.interceptors = filt_int


There are lots of ways to acquire Big Data with which to fill up a Hadoop cluster, but many of those data sources arrive as fast-moving streams of data. Fortunately, the Hadoop ecosystem contains a component specifically designed for transporting and writing these streams: Apache Flume. Flume provides a robust, self-contained application which ensures reliable transportation of streaming data. Flume agents are easy to configure, requiring only a property file and an agent name. Moreover, Flume’s simple source-channel-sink design allows us to build complicated flows using only a set of Flume agents. So, while we don’t often address the process of acquiring Big Data for our Hadoop clusters, doing so is as easy and fun as taking a log ride.

NameNode, Secondary NameNode and Safe Mode – Hadoop

December 26, 2015 Leave a comment

Hadoop Architecture is similar to the master-slave architecture. The master node is called the namenode and the slave nodes are called datanodes. The datanodes are also called as worker nodes. The namenode is the heart of the hadoop system and it manages the filesystem namespace. The namenode stores the directory, files and file to block mapping metadata on the local disk. The namenode stores this metadata in two files, the namespace image and the edit log.

Safe Mode

When you installed the hadoop, the image and edit log files will be empty. As soon as the clients starts writing the data, the namenode writes the file related metadata to the edit log and constructs the file to block and block to datanode mapping in the memory. The namenode does not write the block to data node mapping persistently on the local disk. It just maintains this mapping in the memory. When you restart the namenode, then the namenode goes into a safemode. In this safemode, the namenode merges all the edit log files and writes the metadata into the image file. The namenode then clears the edit log files and waits for the datanodes to reports the blocks. As soon as the datanodes starts reporting the blocks, the namenode constructs the block to datanode mapping in the memory. Once the datanodes completes reporting or cross a certain threshold, the namenode comes out of the safe mode. In safe mode the client cannot do any operations.

As the namenode writes the metadata into the memory, the number of files that you can store in hadoop is controlled by the namenode memory size. As a thumb rule, if the metadata of file takes 150 bytes, then to store one million files you need at least 300MB of ram on the namenode.

Without the namenode, you cannot access the hadoop system. If the namenode goes down then the complete hadoop system is inaccessible. So it is always better to take backup of the image and edit log files. Hadoop provides two ways for taking backup of the image and edit log files. One way is to use the secondary namenode and the second way is to write the files to a remote system.

Secondary NameNode

The secondary Namenode job is to periodically merge the image file with edit log file. This merge operation is to avoid the edit log file from growing into a large file. The secondary namenode runs on a separate machine and copies the image and edit log files periodically. The secondary namenode requires as much memory as the primary namenode. If the namenode crashes, then you can use the copied image and edit log files from secondary namenode and bring the primary namenode up. However, the state of secondary namenode lags from the primary namenode. So in case of namenode failure, the data loss is obvious.

Namenode and secondary namenode

Note: You cannot make the secondary namenode as the primary namenode.

Remote File System

You can configure the namenode to write the image and edit log files to multiple remote network file systems. These write operations are synchronous and atomic. The namenode first writes the metadata to the edit log on the local disk, then writes to the remote NFS system and then to the memory. Once all these operations are success, then the namenode commits the state. In case of any failure, you can copy the files from the remote NFS system and start the namenode. In this case, there is no loss of data.

Big Data Hadoop Architecture and Components

December 26, 2015 Leave a comment

Hadoop is an apache open source software (java framework) which runs on a cluster of commodity machines. Hadoop provides both distributed storage and distributed processing of very large data sets. Hadoop is capable of processing big data of sizes ranging from Gigabytes to Petabytes.

Hadoop architecture is similar to master/slave architecture. The architecture of hadoop is shown in the below diagram:

Hadoop MRV1 Architecture:


Hadoop Architecture

Hadoop Architecture Overview:

Hadoop is a master/ slave architecture. The master being the namenode and slaves are datanodes. The namenode controls the access to the data by clients. The datanodes manage the storage of data on the nodes that are running on. Hadoop splits the file into one or more blocks and these blocks are stored in the datanodes. Each data block is replicated to 3 different datanodes to provide high availability of the hadoop system. The block replication factor is configurable.

Hadoop Components:

The major components of hadoop are:

  • Hadoop Distributed File System: HDFS is designed to run on commodity machines which are of low cost hardware. The distributed data is stored in the HDFS file system. HDFS is highly fault tolerant and provides high throughput access to the applications that require big data.
  • Namenode: Namenode is the heart of the hadoop system. The namenode manages the file system namespace. It stores the metadata information of the data blocks. This metadata is stored permanently on to local disk in the form of namespace image and edit log file. The namenode also knows the location of the data blocks on the data node. However the namenode does not store this information persistently. The namenode creates the block to datanode mapping when it is restarted. If the namenode crashes, then the entire hadoop system goes down.
  • Secondary Namenode: The responsibility of secondary name node is to periodically copy and merge the namespace image and edit log. In case if the name node crashes, then the namespace image stored in secondary namenode can be used to restart the namenode.
  • DataNode: It stores the blocks of data and retrieves them. The datanodes also reports the blocks information to the namenode periodically.
  • JobTracker: JobTracker responsibility is to schedule the clients jobs. Job tracker creates map and reduce tasks and schedules them to run on the datanodes (tasktrackers). Job Tracker also checks for any failed tasks and reschedules the failed tasks on another datanode. Jobtracker can be run on the namenode or a separate node.
  • TaskTracker: Tasktracker runs on the datanodes. Task trackers responsibility is to run the the map or reduce tasks assigned by the namenode and to report the status of the tasks to the namenode.