CarbonData is a fully indexed columnar and Hadoop native data-store for processing heavy analytical workloads and detailed queries on big data with Spark SQL. CarbonData allows faster interactive queries over PetaBytes of data.
CarbonData has specially engineered optimizations like multi level indexing, compression and encoding techniques targeted to improve performance of analytical queries which can include filters, aggregation and distinct counts where users expect sub second response time for queries on TB level data on commodity hardware clusters with just a few nodes.
Unique data organisation for faster retrievals and minimise amount of data retrieved
Advanced push down optimisations for deep integration with Spark so as to improvise the Spark DataSource API and other experimental features thereby ensure computing is performed close to the data to minimise amount of data read, processed, converted and transmitted(shuffled)
Multi level indexing to efficiently prune the files and data to be scanned and hence reduce I/O scans and CPU processing
CarbonData has rich set of features to support various use cases in Big Data analytics. The below table lists the major features supported by CarbonData.
CarbonData provides its own DDL to create and manage carbondata tables. These DDL conform to Hive,Spark SQL format and support additional properties and configuration to take advantages of CarbonData functionalities.
CarbonData provides its own DML to manage data in carbondata tables.It adds many customizations through configurations to completely customize the behavior as per user requirement scenarios.
CarbonData supports Update and Delete on Big Data.CarbonData provides the syntax similar to Hive to support IUD operations on CarbonData tables.
CarbonData has unique concept of segments to manage incremental loads to CarbonData tables effectively.Segment management helps to easily control the table, perform easy retention, and is also used to provide transaction capability for operations being performed.
CarbonData supports 2 kinds of partitions.1.partition similar to hive partition.2.CarbonData partition supporting hash,list,range partitioning.
CarbonData manages incremental loads as segments. Compaction helps to compact the growing number of segments and also to improve query filter pruning.
CarbonData can read any carbondata file and automatically infer schema from the file and provide a relational table view to perform sql queries using Spark or any other applicaion.
CarbonData supports bloom filter index in order to quickly and efficiently prune the data for scanning and acheive faster query performance.
Lucene is popular for indexing text data which are long.CarbonData supports lucene index so that text columns can be indexed using lucene and use the index result for efficient pruning of data to be retrieved during query.
MVs are kind of pre-aggregate and pre-join tables which can support efficient query re-write and processing.CarbonData provides MV which can rewrite query to fetch from any table(including non-carbondata tables). Typical usecase is to store the aggregated data of a non-carbondata fact table into carbondata and use mv to rewrite the query to fetch from carbondata.
CarbonData supports streaming of data into carbondata in near-realtime and make it immediately available for query.CarbonData provides a DSL to create source and sink tables easily without the need for the user to write his application.
CarbonData supports writing data from non-spark application using SDK.Users can use SDK to generate carbondata files from custom applications. Typical usecase is to write the streaming application plugged in to kafka and use carbondata as sink(target) table for storing.
CarbonData supports reading of data from non-spark application using SDK. Users can use the SDK to read the carbondata files from their application and do custom processing.
CarbonData can write to S3, OBS or any cloud storage confirming to S3 protocol. CarbonData uses the HDFS api to write to cloud object stores.
CarbonData uses HDFS api to write and read data from HDFS. CarbonData can take advantage of the locality information to efficiently suggest spark to run tasks near to the data.
CarbonData also supports read and write with Alluxio.
CarbonData is useful in various analytical work loads.Some of the most typical usecases where CarbonData is being used is documented here.