Hadoop architecture by ajay

http://www.beinghadoop.com

hadoopframework@gmail.com

MASTER NODE, SLAVE NODES.
Master nodes are
Namenode
Secondary namenode
Jobtracker
Slave nodes:
Datanodes, Task trackers.
Hdfs – Hadoop distributed file system which is distributed to
Available Datanodes. So actual data resides in Datanodes.
Data in Datanodes represented as small size chunks,
we call this chunks as blocks.
The default block size in HDFS is 64Mb.
For example if you want to store 1GB of data, in HDFS,
In 3 data nodes, 1000mb/64mb=15.625mb
16 blocks are equally distributed 3 data nodes.
In This 15 blocks are in same size, 16th one with 7Mb.

These blocks are replicated based on replication policy.
Data replication policy is to improve data reliability,
availability and network bandwidth utilization.

If you are using Replication of blocks, the Default Replication
Factor is 3. In above example 3*16=48 blocks are created.
These 48 blocks are replicated to all the data nodes
As HDFS replication placement policy
Hdfs stores --one replica on node in local rack.
second replica on another datanode in local rack,
third one in different node in a different rack.
Hdfs stores each file as sequence of blocks.
All blocks of a file except last block are the same size.
this blocks of a file are replicated for fault tolerance.
A user or an application can specify the replication factor for a file.

The default Replication factor for HDFS is specified
In Hdfs-site.xml under hadoop-home/conf directory

We can manually specify replication factor for file using
Hadoop fs –setrep command.
Syntax: hadoop fs s-setrep [-R] location of file.
R indicates Recursive


In above diagram data1 contains 3 blocks 1,2,3
Data2 contains 2 blocks 4,5. There are 4 data nodes.
This 5 blocks*3 replications=15blocks are distributed to
4 data nodes.
When blocks are replicating to data nodes, General issues arises like
---over replicated blocks
---under replicated blocks
---corrupted blocks
---miss replicated blocks.

To balance These Nodes equally
We run a script start-balancer.sh under Hadoop-home/bin directory
We use
hadoop fsck /,
hadoop dfsadmin –report
hadoop dfsadmin –metasave metaloginfo.txt
Datanode blockscanerreport
To check blocks information, Storage information
health of hdfs file system,


We can not access the data from Datanode Directly.
Because datanode contains distributed blocks,
Which are like fragmented data or part files.
The Datanode are responsible for
---serving read, write requests from the clients
---performs block creating, deletion, Replication of blocks based on replication factor..


when datanode startup, each datanode in a cluster performs
handshake procedure with namenode.
Each DataNode sends its block report to the NameNode every hour
so that NameNode has an up to date view of where block replicas are
located in the cluster.

Datanode also sends heartbeats to the NameNode every ten minutes.
so that Namenode will identify how many datanodes are working properly.
if it doesn't receive a heartbeat from DataNode,
NameNode assumes that the particular datanode got failure.
And then it starts writing replicas in available datanodes.


Name node manages the file system namespace.
It stores metadata for the files which are being stored in DataNode.
NameNode maintains the data in the form file system tree.
When user writes data to DataNode, its metadata is maintained in Namenode.
So when user tries to access the same data, or Hdfs client requests to access data,
it has to contact NameNode through Jobtracker. It gets the reference in which
DataNode, the data is located. Will be given to Hdfs client.
When the Namenode starts up it gets the
FsImage and Editlog from its local file system,
update FsImage with EditLog information and then
stores a copy of the FsImage on the filesytstem as a checkpoint
When user initiated write operation, the data written to editlog file.
When chekpoint occurs it writes the data into fsimage file

Name node stores the data as a Hierarchical file system
Namenode maintains the file system tree
Any meta information changes to the file system
is recorded by the Namenode.
An application can specify the
number of replicas of the file needed,
replication factor of the file.
This information is stored in the Namenode.

Namenode uses a transaction log called the EditLog to record
every change that occurs to the filesystem meta data.
For example, creating a new file.
Change replication factor of a file
EditLog is stored in the Namenode’s local filesystem
Entire filesystem namespace including mapping of blocks to
files and file system properties is stored in a file FsImage. Stored
in NameNode’s local file system.

When you start the Namenode , it enters to Safemode.
During the safemode, Replication of data blocks do not occur.
Each DataNode checks in with Heartbeat and BlockReport.

Namenode verifies that each block has
acceptable number of replicas
After a configurable percentage of safely replicated blocks check in
with the Namenode, Namenode exits Safemode.
It then makes the list of blocks that need to be replicated.
Namenode then proceeds to replicate these blocks to other
Datanodes.


Client applications submit jobs to the Job tracker.
The JobTracker talks to the NameNode to determine the location of the data
The JobTracker locates TaskTracker nodes with available slots
at or near the data
The JobTracker submits the work to the chosen TaskTracker nodes.
The TaskTracker nodes are monitored. If they do not submit heartbeat
signals often enough, they are treated as they have failed and the work is
scheduled to a different TaskTracker.
A TaskTracker will notify the JobTracker when a task fails.
The JobTracker decides what to do then:
it may resubmit the job elsewhere,
it may mark that specific record as something to avoid,
it may even blacklist the TaskTracker as unreliable.
When the work is completed, the JobTracker updates its status.
Client applications can poll the JobTracker for information

A Tasktracker is a slave node in the cluster which that accepts the tasks
from JobTracker like Map, Reduce or shuffle operation. Tasktracker also runs in
its own JVM Process.
Every TaskTracker is configured with a set of slots; these indicate the number of
tasks that it can accept. The TaskTracker starts a separate JVM processes to do
the actual work (called as Task Instance) this is to ensure that process failure
does not take down the task tracker.
The Tasktracker monitors these task instances, capturing the output and exit
codes. When the Task instances finish, successfully or not, the task tracker
notifies the JobTracker.
The TaskTrackers also send out heartbeat messages to the JobTracker, usually
every few minutes, to reassure the JobTracker that it is still alive. These
messages also inform the JobTracker of the number of available slots, so the
JobTracker can stay up to date with where in the cluster work can be delegated.

The Maximum Number of Map tasks that can run on a task tracker
Is controlled by the property for map tasks
Mapred.tasktracker.map.tasks.maximum=2(Default 2)
The Maximum Number of Reduce tasks that can run on a task tracker
Is controlled by the property for reduce tasks
Mapred.tasktracker.reduce.tasks.maximum=2 (Default=2)
If eight processors are available we can assign maximum Seven for map
Or reducer tasks.


1. A block is minimum amount of data that can read or Write.
HDFS- The Default block size 64mb.
we can set it to 128mb in hdfs-site.xml
dfs.blocksize=128mb
2. The HDfs default buffer size 4kb.
By considering system configuration we can set
this using io.file.buffer.size in core-site.xml


When we add a node to existing cluster
it is commissioning a node,
Removing a node is decommissioning node.

3.To commissioning a node
Add the network address of the nodes to commissioning.
Dfs.hosts= node address (hdfs-site.xml)

Mapred.hosts=node address(mapred-site.xml)
Update the name node using the commands
Hadoop dfsadmin –refreshnodes
Update the jobtracker using the command
Hadoop mradmin –refreshnodes.

To Decommissioning a node
Add network addresses of the nodes to be decommissioned.
Dfs.hosts.exclude=node address(hdfs-site.xml)
Mapred.hosts.exclude=node address(mapred-site.xml)
Update the name node using the commands
Hadoop dfsadmin –refreshnodes
Update the jobtracker using the command
Hadoop mradmin –refreshnodes

4.Non HDFs in data node:
In One Tb disk of data node, If we want to use
250Gb for Non Hdfs storage
Set dfs.datanode.du.reserverd=250gb(in bytes)


5. Recycle bin for HDFS.
If trash enabled , A hidden directory created with the name .trash
Command to clear the trash for Hdfs
Hadoop fs –expunge.
To specify the minimum amount of time that
a file will remain in trash can be set using
fs.trash.interval=600 minutes.
To disable the trash set
Fs.trash.interval=0 in core.site.xml


6. DATANODE BLOCK SCANNER:
PEROIODICALLY VERIFIES ALL THE BLOCKS ON THE DATA NODE.
The default interval every 504 HOURS.
HDFS-SITE.XML:
DFS.DATANODE.SCAN.PERIOD.HOURS=504

7. Log Files Location: By default The log directory is located
Under hadoop_install/logs. We can assign new location for
Log directory in Hadoop-env.sh by adding a Line
Export Hadoop_log_dir=/var/log/hadoop


Hadoop-env.sh:
Environmental variables which are used to run hadoop.

Core-site.xml: contains configuration settings for hadoop, i/o settings,
other properties common to Hdfs and Mapreduce.
Hdfs-site.xml:
Configurations for hdfs daemons,
Namenode, snn, data nodes.
Mapred-site.xml:
Configuration settings for Mapreduce daemons:
Jobtracker, tasktrackers.


To designate particular node as name node, specify its URI
In core-site.xml for the property fs.default.name.
The NameNode ‘s name or its IP address is specified
In core-site.xml:

In case of psuedo distributed mode, we configure name node in localhost.
We can also specify like
hdfs://0.0.0.0:50070
http://127.0.0.1:50070

Dfs.name.dir property in hdfs-site.xml contains the location
Where NameNode stores file system metadata. We can specify multiple
Disks locations, remote disks. So that we can recover data incase NameNode
Got failure. Specify the directory names separated by comma(,).


The property dfs.data.dir in hdfs-site.xml specify the location
Where Datanode stores its blocks. This property contains list of directories
Where DataNode stores its blocks. We can specify multiple directories for replication
Of blocks. It uses round robin algorithm to write data between directories.

Mapred.job.tracker property in mapred-site.xml specifies
The port where job tracker is running.
In a fully distributed mode we can configure jobtracker in a separate node too.


Mapred.local.dir which contains list of directory name seperated by comma
Where jobtracker stores intermediate data for jobs. The data is cleared when
Job is completed.

Mapred.system.dir specifies a locations where shared files are stored
While a job is running.
Mapred.tasktracker.map.tasks.maximum is a property specifies
The number of map tasks that can run on a tasktracker at one time.
Mapred.tasktracker.reduce.tasks.maximum is a property specifies
The number of reducer tasks that can run on a tasktracker at one time.


Name node’s http server address port: 50070.
Open web browser and type uri for NameNode as
http://localhost:50070
http://0.0.0.0:50070
Http://127.0.0.1:50070

50030 ; jobtracker http server address port
50060: task tracker http server address port
50075: datanode’s http server address port
50090: secondary name node’s http server address port


These Scripts are located under the Directory
Hadoop_home/bin or
/usr/lib/bin
HADOOP CONTROL SCRIPTS:
Master file: It is plain text file Contains the
NameNode, Secondary NameNode, Jobtracker address.
Slave file. It is a plain text file contains the address of data nodes.
Start-dfs.sh:
Starts namenode on the local machine.
Starts datanode on each machine listed in slave file.
Starts secondary namenode .
Start-mapred.sh
Starts a jobtracker
Starts tasktracker on each datanode machine.
To stop this we have the scripts
Stop-dfs.sh
Stop-mapred.sh

http://0.0.0.0:50075
Http://127.0.0.1:50075


http://0.0.0.0:50030
Http://127.0.0.1:50030

http://0.0.0.0:50030
Http://127.0.0.1:500


http://0.0.0.0:50090
Http://127.0.0.1:50090


http://0.0.0.0:50060
Http://127.0.0.1:50060


Hadoop architecture by ajay

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (20)

Destacado

Destacado (7)

Similar a Hadoop architecture by ajay

Similar a Hadoop architecture by ajay (20)

Último

Último (20)

Hadoop architecture by ajay