Getting Data into Hadoop
- Hadoop as a Data Lake
- The Hadoop Distributed File System (HDFS)
- Direct File Transfer to Hadoop HDFS
- Importing Data from Files into Hive Tables
- Importing Data into Hive Tables Using Spark
- Using Apache Sqoop to Acquire Relational Data
- Using Apache Flume to Acquire Data Streams
- Manage Hadoop Work and Data Flows with Apache Oozie
- Apache Falcon
- What's Next in Data Ingestion?
Hortonworks data scientists focus on data ingestion, discussing various tools and techniques to import datasets from external sources into Hadoop. They begin with describing the Hadoop data lake concept and then move into the various ways data can be used by the Hadoop platform. The ingestion targets two of the more popular Hadoop tools—Hive and Spark.
You can have data without information, but you cannot have information without data.
Daniel Keys Moran
No matter what kind of data needs processing, there is often a tool for importing such data from or exporting such data into the Hadoop Distributed File System (HDFS). Once stored in HDFS the data may be processed by any number of tools available in the Hadoop ecosystem.
This chapter begins with the concept of the Hadoop data lake and then follows with a general overview of each of the main tools for data ingestion into Hadoop—Spark, Sqoop, and Flume—along with some specific usage examples. Workflow tools such as Oozie and Falcon are presented as tools that aid in managing the ingestion process.
Hadoop as a Data Lake
Data is ubiquitous, but that does not always mean that it’s easy to store and access. In fact, many existing pre-Hadoop data architectures tend to be rather strict and therefore difficult to work with and make changes to. The data lake concept changes all that.
So what is a data lake?
With the more traditional database or data warehouse approach, adding data to the database requires data to be transformed into a pre-determined schema before it can be loaded into the database. This step is often called “extract, transform, and load” (ETL) and often consumes a lot of time, effort, and expense before the data can be used for downstream applications. More importantly, decisions about how the data will be used must be made during the ETL step, and later changes are costly. In addition, data are often discarded in the ETL step because they do not fit into the data schema or are deemed un-needed or not valuable for downstream applications.
One of the basic features of Hadoop is a central storage space for all data in the Hadoop Distributed File Systems (HDFS), which make possible inexpensive and redundant storage of large datasets at a much lower cost than traditional systems.
This enables the Hadoop data lake approach, wherein all data are often stored in raw format, and what looks like the ETL step is performed when the data are processed by Hadoop applications. This approach, also known as schema on read, enables programmers and users to enforce a structure to suit their needs when they access data. The traditional data warehouse approach, also known as schema on write, requires more upfront design and assumptions about how the data will eventually be used.
For data science purposes, the capability to keep all the data in raw format is extremely beneficial since often it is not clear up front which data items may be valuable to a given data science goal.
With respect to big data, the data lake offers three advantages over a more traditional approach:
All data are available. There is no need to make any assumptions about future data use.
All data are sharable. Multiple business units or researchers can use all available data1, some of which may not have been previously available due to data compartmentalization on disparate systems.
All access methods are available. Any processing engine (MapReduce, Tez, Spark) or application (Hive, Spark-SQL, Pig) can be used to examine the data and process it as needed.
To be clear, data warehouses are valuable business tools, and Hadoop is designed to complement them, not replace them. Nonetheless, the traditional data warehouse technology was developed before the data lake began to fill with such large quantities of data. The growth of new data from disparate sources including social media, click streams, sensor data, and others is such that we are starting to quickly fill the data lake. Traditional ETL stages may not be able to keep up with the rate at which data are entering the lake. There will be overlap, and each tool will address the need for which it was designed.
The difference between a traditional data warehouse and Hadoop is depicted in Figure 4.1.
Figure 4.1 The data warehouse versus the Hadoop data lake.
Different data sources (A, B, C) can be seen entering either an ETL process or a data lake. The ETL process places the data in a schema as it stores (writes) the data to the relational database. The data lake stores the data in raw form. When a Hadoop application uses the data, the schema is applied to data as they are read from the lake. Note that the ETL step often discards some data as part of the process. In both cases the user accesses the data they need. However, in the Hadoop case it can happen as soon as the data are available in the lake.