BigData Ecosystem Architecture

Internal working of Bigdata and it's ecosystems such as

• The background process of resource allocation, database connection.
• How the data is distributed across the nodes.
• Execution life-cycle on submitting a Job.

** Note: Refer the links metioned below under each ecosystem for detailed explanation **

1. HDFS :elephant:

The various underlying process that takes place during the storage of a file into HDFS such as:

• Type of scheduler

• Block & Rack information

• File size

• File location

• Replication information about the file(Over-replicated blocks, Under-replicated blocks, ...)

• Health status of the file

Please click on the link below to know the execution and flow process

:link: HDFS Architecture in Depth

2. SQOOP :octocat:

Used to perform 2 main operations.

• Sqoop Import:

  • To ingest data from any source such as traditional databases into hadoop file system HDFS

• Sqoop Export:

  • To export data from hadoop file system HDFS to any traditional databases

To support the above two operations internally a CodeGen is used.

• Sqoop CodeGen:

  • To compile metadata and other relative information into java class file & create a Jar

Please click on the link below to know the execution and flow process

:link: SQOOP Architecture in Depth

3. HIVE :honeybee:

It has mainly 4 components

• Hadoop core components(Hdfs, MapReduce)

• Metastore

• Driver

• Hive Clients

Please click on the link below to know the execution and flow process

:link: HIVE Architecture in Depth

4. SPARK :boom:

The various phases involved before and during the execution of a spark job.

• Spark Context

  • It is the heart of spark application.

• Yarn Resource Manager, Application Master & launching of executors (containers).

• Setting up environment variables, job resources.

• CoarseGrainedExecutorBackend & Netty-based RPC.

• SparkListeners.

  • LiveListenerBus
  • StatsReportListener
  • EventLoggingListener

• Execution of a job

  • Logical Plan (Lineage)
  • Physical Plan (DAG)

• Spark-WebUI.

Please click on the link below to know the execution and flow process

:link: SPARK Architecture in Depth

4.1 SPARK Abstraction Layers & Internal Optimization Techniques used :boom:

It has 3 different variants as part of it.

• RDD (Resilient Distributed Datasets)

  • Lineage Graph
  • DAG Scheduler

• DataFrames

  • Catalyst Optimizer
  • Tungsten Engine
  • Default source or Base relation

• Datasets

  • Optimized Tungsten Engine - V2
  • Whole Stage Code Generation

5. HBASE :whale2: